238Hybrid structural topology of the proposed methodequation.3.

23 δ10 268Hybridstructuraltopologyof theproposedmetho

1
 Sructural Topology Patten for Texture Classification

October 9, 2019

Abstract
The local descriptor adds interesting features to the classification of texture images and has gained in
recent years a lot of interest in many applications. Such as texture classification, face recognition, image
retrieval, so and on. This paper proposes a novel local feature descriptor named: Structural Topology
Pattern (STP) for texture classification. The proposed operator obtained after concatenations of the four
parametric methods that do not only take into account the difference in grey value between the central pixels
and the adjacent pixels in the eight directions, but it compares at three levels the values of the intensities
of the image exclusively, for the first two descriptors (upper ,lower):
Level 1: Comparison of the grey values of the four pairs of symmetrical pixels to the central pixel,
Level 2: comparison of the average grey values of the pixels of level 1,
Level 3: Compares the average of local and global pixel values for the entire image.
in the same way for the last two descriptors (upper, lower). but this time they exclusively compare the pixel
values of the two-level image:
Level 1: comparison of the grey values of the four pixels of the odd directions,
Level 2: comparison of the grey values of the four pixels of the odd directions.
The texture descriptors are compared and evaluated on twelve texture datasets; Brodatz, KTH-TIPS, KTH-
TIPS2b, Kylberg, Jerry Wu, XUH, CUReT, USPtex1.0, New BarkTex, Outex-TC13, VisTex and RSMAS
texture dataset. After optimizing all parameters for each dataset. they have experimentally proven to be
more accurate than most of the most recent texture classification methods. The advantages of the proposed
texture analysis descriptors include low computational complexity, rotation invariant, low impulse-noise
sensitivity and high usability are advantages of the

Stuctural Topology Pattern (STP),Optimization parameters, Feature extraction, Texture classification

1 1 Introduction
2 Texture classification is an important method developed in recent years and widely used in many computer
3 vision and image analysis applications,such as medical image analysis [1], remote sensing[2], facial recognition
4 [3] and image retrieval [4], etc., in this regard, many research and studies have been launched in the same
5 context, without achieving a universal definition . An image texture was defined as the local spatial variations
6 in pixel intensities and orientation [5] The recent textbook by Pietikäinen et al. [6] provide a good description
7 of texture in stating that “A textured area in an image can be characterized by a non-uniform or varying spatial
8 distribution of intensity or color”.The texture defined by Haralick et al [7] describes the spatial distribution of
9 the gray levels of an [Link] the same context, the extraction of the characteristics of the texture analysis
10 were classified by X. Xie[8], in four categories, namely statistical approaches,structural approaches, filter based
11 methods, and model based approaches. Table 1 shows a summary list of some of the key texture analysis
12 methods that have been applied to Texture classification or segmentation. Clearly, statistical and filter based
13 approaches have been very [Link] of the interests of Local Binary Pattern (LBP) introduced by Ojala et
14 al[9] as a descriptor that generates a number of texture units more restricted for a neighborhood of eight pixels
15 of an matrix (3 × 3) , this method is successfully used in various fields, in terms of discriminating efficiency,
16 and performance it also contains structural and statistical information,it is also invariant to rotations and scale
17 changes and offers a high tolerance to changes in brightness. this method is also of low complexity which allows
18 2D images to be analyzed in real time. noise sensitivity and low accuracy remain one of the most important
19 issues facing this operator. In the same context, several research projects have been launched. containing
20 different methods converging towards discriminatory results. Including parametric methods (made up of one or
21 more parameters)
22

23 Including Local Ternary Pattern LTP [10], Improved Local Ternary Pattern ILTP [11] and Local Quinary
24 Patterns LQP [12] etc. which are part of the improved types of LBP. Both use one threshold and The last
25 descriptor uses two different thresholds. Namely that these thresholds are determined by the user, a three-digit
26 coding is created for the LTP and ILTP descriptors, then using a function, this pattern is divided into a ternary
 Figure 1: Inexhaustive list of textural analysis methods

27 local, likewise a four-digit coding has been created for the descriptor LQP ,so as its function is divided into four
28 local quinary patterns. Finally the histograms obtained by these operators from the above binary patterns are
29 concatenated to form the final descriptor. In this paper, we present a new method called the Structural Topology
30 Pattern with notation [Link] this paper, we present a new method called structural modeling with the STP
31 [Link] parametric methods (made up of one or more parameters).Including Local Ternary Pattern
32 LTP [x1], Improved Local Ternary Pattern ILTP and Local Quinary Patterns LQP etc,which are part of the
33 improved types of [Link] first two descriptors use one threshold for each, while the latter uses two different
34 [Link] that these thresholds are determined by the user, A three-digit encryption is created to
35 describe LTP and ILTP, then using a function, this pattern is divided into three binary patterns , and similarly,
36 for the LQP operator, a five-digit encryption is created, and then using a function, this pattern is divided into
37 four binary patterns. Finally, the histograms obtained by these operators from the above binary patterns are
38 concatenated to form the
39 final [Link] this paper, we present a new method called structural topology pattern with the STP
40 notation. A major disadvantage of parametric methods LTP and [Link] depend on comparing the values
41 of adjacent pixels with the values obtained by subtracting the threshold from the value of the central pixel or
42 adding the threshold to the value of the central [Link] to the methods described above. Moreover,
43 the precisions of the analysis of these operators are very sensitive to this structural [Link] approaches
44 depend on two [Link] first approach is based on a mixture of operations for the values of the pixels
45 adjacent to the central pixel in their local or global context and then comparing them to the value of the
46 local central pixel with the addition of the first threshold or subtracting it as is the case for the LTP operator
47 methods as explained [Link] second approach also depends on a different mixture than the one referred to
48 in the first approach It consists of several operations of adjacent pixel values in their local and global context
49 with the addition of the second threshold or subtracting it as is the case for the LTP operator methods as
50 explained [Link] on the two approaches, we get two operators with different thresholds instead of two
51 different threshold operators,as is the case with the LQP [Link] order to solve the above problems, the new
52 descriptor STP is composed of four descriptors inspired by the LTP descriptor, so that the first two descriptors
53 which are the upper descriptor ST P (Uτ 1 ) and the lower descriptor ST P (Lτ 1 ), menu of a threshold τ1 , while
54 the last two are the upper descriptor ST P (Uτ 2 ) and the lower descriptor ST P (Lτ 2 ) also menu of a threshold
55 τ2 , where τ1 and τ2 are positive parameters defined by the user.

3
56 2 Revisiting the existing methods
57 In this section, we present a brief overview of LBP, LTP,ILTP and LQP. Figure 1 shows an example of 8
neighboring pixels around the central pixel.

Figure 2: An example of 8 neighboring pixels around the central pixel.

58

59 2.1 local binary pattern (LBP)

60 The LBP operator was first mentioned in 1993 to measure the local contrast of an image, but was later pop-
61 ularized by (Ojala et al., 2002)[9] to analyze [Link], the neighbors of each pixel are defined as a set of
62 sampling points (ap ) around a central pixel (ac ), and then the luminance level of the central pixel is compared
63 to the adjacent pixel levels. The LBP descriptor is defined as follows:
7
X
LBP = 2p ξ(ap − ac ) (1)
p=0
64 (
1 if x >= 0,
ξ(x) = (2)
0 else
65 where ap is the gray value of the neighbors of the central pixel ac

66 2.2 Local Ternary Pattern (LTP)

67 For the LBP operator, a slight change in gray of the central pixel may result in different codes for a neighborhood
68 in an image, which means that this descriptor is sensitive to noise. To fix this problem Tan and Triggs [10]
69 extended the basic LBP to a version with three-value codes, which is called the local ternary pattern (LTP).
70 In LTP, the indicator ξ(·) ( Eq. (2)) is replaced by 3-valued function ϕ(·, ·, ·) ( Eq. (3) and binary LBP code is
71 replaced by ternary LTP code as follows.

+1 ap > ac + τ ,

ϕ(ap , ac , τ ) = 0 |ap − ac | < τ , (3)

−1 ap 6 ac − τ ,


72 where τ is a threshold specified by the user.

73 Consequently, by using the function ϕ(ac , ac , τ ),In order to reduce the size of the functionality, a coding scheme
74 is also represented by Tan and Triggs [31] by splitting each ternary pattern into two parts:local ternary patterns

4
 Figure 3: Calculation of the LBP with eight neighboring pixels.

75 lower (LT PL ) and local ternary patterns upper (LT PU ) as illustrated in Figure 2. are coded as following ( Eq.
76 (4) and Eq. (5):
X7
LT PL = 2p ξ(ac − ap − τ ) (4)
p=0
77
7
X
LT PU = 2p ξ(ap − ac − τ ) (5)
p=0

78 The two descriptor histograms hLT PL and hLT PU are finally concatenated to form the final hLT P future vector
79 as illustrated in Eq. (6). LTP generates 2.28 possible different patterns.

hLT P = hLT PL ||hLT PU (6)

80 2.3 Improved Local Ternary Pattern (ILTP)

81 Improved local ternary patterns (ILTP) [11] are an extension of LTP where each pixel in the neighbourhood is
82 thresholded at the average grey-scale value. Similarly to LTP the representation is split into a lower and upper
83 part:
7
X
ILT PL = 28 ξ(S̄ − ac − τ ) + 2p ξ(S̄ − ap − τ ) (7)
p=0

7
X
ILT PU = 28 ξ(ac − S̄ − τ ) + 2p ξ(ap − S̄ − τ ) (8)
p=0

1 P7
84 Where {S = (ac + p=0 ap ) is the average value of the whole 3 × 3 neighbourhood . The two descriptors
9
85 are concatenated to give the ILTP model:

hILT P = hILT PL ||hILT PU (9)

5
 Figure 4: Calculation of the LTP with eight neighboring pixels

86 2.4 Local Quinary Patterns (LQP)

87 In local quinary patterns (LQP) [12], the difference of the gray level value between the central pixel ac and the
88 gray levels of the neighboring pixels ap is encoded according to five levels (i.e. 2, -1, 0, 1 and 2) calculated using
89 two thresholds: τ1 and τ2 . The LQP is therefore closely related to the LTP, the only difference being that the
90 number of coding levels is five in the LQP while in the LTP it is three. The quinary model is divided into four
91 binary models according to the following rule:



+2 ap > ac + τ2 ,
+1 ac + τ1 6 ap < ac + τ2 ,



ϕ(ap , ac , τ1 , τ2 ) = 0 ac − τ1 6 ap < ac + τ1 , (10)

−1 ac − τ2 6 ap < ac + τ1 ,





−2 Otherwise ,

92 where τ1 and τ2 are two thresholds specified by the user.

93 As a result, using the ϕ(ap , ac , τ1 , τ2 ), to reduce the size of the feature, splitting each quinary pattern into four
94 parts: (LQP−2 ), (LQP−1 ), (LQP+1 ) and (LQP+2 ) as shown in figure 3. are coded as follows (Eq. (11),Eq.
95 (12),Eq. (13) and Eq. (14):
7
X
LQP−2 = 2p ξ(ac − ap − τ2 ) (11)
p=0
96
7
X
LQP−1 = 2p ξ(ap − ac + τ2 )ξ(ac − ap + τ1 ) (12)
p=0
97
7
X
LQP+1 = 2p ξ(ap − ac − τ1 )ξ(ac − ap + τ2 (13)
p=0
98
7
X
LQP+2 = 2p ξ(ap − ac − τ2 ) (14)
p=0

6
 99 The four descriptor histograms (hLQP−2 ), (hLQP−1 ), (hLQP+1 ) and (hLQP+2 ) are finally concatenated to
100 form the final hLQP future vector as illustrated in Eq. (15). LQP generates 4.28 possible different patterns.

LQP = hLQP−2 ||hLQP−1 ||hLQP+1 ||hLQP+2 (15)

101 3 Hybrid structural topology of the proposed method

102 LetIM ×N =(ai,j ), be an image of order M × N where 0 ≤ i ≤ M − 1 and 0 ≤ j ≤ N − 1,for practical reasons,
103 we define the 3 × 3, square neighborhood used in this study. For this, we assume that 3 × 3, ([Link].(16)) as a
104 local image of order 3 × 3 of the global image IM ×N consisting of gray scale values of a neighborhood centered
105 on the pixel coordinates (i, j):
106

107  
ai−1,j−1 ai−1,j ai−1,j+1
I3×3 =  ai,j−1 ai,j ai,j+1  (16)
ai+1,j−1 ai+1,j ai+1,j+1
108 to simplify the computational complexity caused by the indices i and j of neighboring gray scale values I3 ×3(i, j),
109 we simply define I3 × 3 as indicated by ([Link](2))
 
a3 a2 a1
I3×3 =  a4 ac a0  (17)
a5 a6 a7
110

111 From the global image IM ×N (macrostructure) and the local image I3×3 (micro-structure), we propose four
112 topological properties (structural and hybrid) to describe the weighted pixel detailed in the following steps:
113 •Step 1 : we note respectively MG , mL and Dk ,the average of the pixels of the global image IM ×N , the average
114 of the pixels of the local image I3×3 and the averages of the pixels according to the directions (kπ) 4 where
k = 0, . . . , 3 as shown in ([Link](3)),([Link].(18)),([Link](19)) and ([Link](20))

(kπ)
Figure 5: Calculat the averages of the pixels according to the directions 4 | k = 0, ...3
115

7
1 X
mL = (ac + ap ) (18)
9 p=0
116
M −1 N −1
1 X X
MG = a(i,j) (19)
M × N i=0 j=0
117
(kπ) (ak + ac + ak+4 )
Derction : Dk = ∀k ∈ {0, 1, 2, 3} (20)
4 3
118 We define two sets Uτ1 ([Link](21)) and Lτ1 ([Link](22)) based on two directions (Upper, Lower) according to a
119 metric topology to compare the central pixle is these neighbors in a structural and hybrid context in which we

7
120 introduce a threshold τ1 defined by the user according to the results experimental

U (τ1 ) = {(x, y) ∈ G1 | (x > ac + τ 1 )and (y > ac − τ1 )} (21)

121

L(τ1 ) = {(x, y) ∈ G1 | (x 6 ac − τ 1 ) and (y 6 ac + τ1 )} (22)

122


 1, if (x, y) ∈ U (τ1 )
δ0 (x, y)) = −1, if (x, y) ∈ L(τ1 ) (23)
0, otherwise


123

124 Where G1 = {(mL , MG ), (D0 , D1 ), (D2 , D3 ), (ak , ak+4 )}, k ∈ {0, ..., 3}; and τ1 , is a positive parameter (thresh-
125 old)
126

127 •Step 2 : Instead of using a code with base 3 to encode all three, we will proceed exactly as the case of
128 the LTP operator which uses two binary codes representing upper and lower components of the ternary code,
129 namely: two binary codes encoding the two {1, −1} the binary codes are grouped in two distinct histograms
130 as indicated by the equations (Eq.28) and (Eq.29) and, in the last step, the histograms are concatenated to
131 form the entity Vector mentioned in equation ([Link].(30), the codes of the first two upper and lower descriptors
132 named structural topology patterns (ST P (Lτ1 ) , ST P (U τ1 ) are given in equations ([Link].(24)and ([Link].(25)
133 as follows:
3
X
ST PU τ 1 (xc , yc ) = δ1 (mL , MG ))26 + δ1 (D0 , D1 ))25 + δ1 (D2 , D3 ))24 + 2k δ1 (ak , ak+4 ) (24)
k=0
134
3
X
ST PLτ 1 (xc , yc ) = δ2 (mL , MG ))26 + δ2 (D0 , D1 ))25 + δ2 (D2 , D3 ))24 + 2k δ2 (ak , ak+4 ) (25)
k=0

135 Where δ1 (., .) and δ2 (., .) are two binary thresholding functions in this case are:

136


1, if (x, y) ∈ Uτ 1
δ1 (x, y)) = (26)
0, otherwise
137


1, if (x, y) ∈ Lτ1
δ2 (x, y)) = (27)
0, otherwise
138

139

140 Where Uτ 1 and Lτ 1 represent the two sets defined in the equations([Link].(21)and ([Link].(22)
141

142 The two descriptor histograms hST P (U τ1 ) and hST P (Lτ1 ) illustrated in equations([Link].(28)and ([Link].(29)
143 are finally concatenated to form the final hST P (τ1 ) future vector as illustrated in equation ([Link].(30). hST P (τ1 )
144 generates 2.27 possible different patterns
M
X −1 N
X −1
hST P (U τ1 ) (k1 ) = ϕ(ST PU τ 1 (i, j), k1 ) (28)
i=0 j=0
145
M
X −1 N
X −1
hST P (Lτ1 ) (k1 ) = ϕ(ST PLτ 1 (i, j), k1 ) (29)
i=0 j=0

hST P (τ1 ) = hST P (U τ1 ) ||hST P (Lτ1 ) (30)


1, if x = y
146 where: ϕ(x, y) = , k1 ∈ [0, 27 ]
0, otherwise
147

148 •Step 3 : In the same way, we define the two other sets Uτ 2 and Lτ 1 but this time with a threshold τ2 We
149 will choose the pixels a2k+1 in the directions (2k+1)π
4 For each k = 0, 1, 2, 3; then the values of their adjacent
150 pixel a2k and a2k+2 , and we calculate the averages of the gray levels of (a2k , a2k+1 ) and (a2k+1 , a2k+2 ), then

8
 π 3π 5π 7π
Figure 6: Examples of neighborhoods of the pixels of the directions ( 4, 4 4 , 4 )

151 the average of these averages mentioned above as shown in ([Link].4) and ([Link].31) next:
152 So we get the average in the case general as follow

(2k + 1)π (a2k + a2k+1 ) (a2k+1 + a2k+2 ) (a2k + 2a2k+1 + a2k+2 )

Direction : B2k+1 = + = (31)
4 2 2 4
153 where k ∈ {0, 1, 2, 3}
154 We then introduce the medians ImedG and ImedL which are respectively the global median of the image IM ×N
155 and the local median of I3×3 image as well as MG and ImedL defined in ([Link].18)) and ([Link].19) Finally, as
156 in the context of step 1, we establish two sets Uτ 2 and Lτ 2 as shown in ([Link].32) and ([Link].33)

U (τ2 ) = {(x, y) ∈ G2 | (x > ac + τ 2 )and (y > ac − τ2 )} (32)

157

L(τ2 ) = {(x, y) ∈ G2 | (x 6 ac − τ 2 ) and (y 6 ac + τ2 )} (33)

158

 1, if (x, y) ∈ U (τ1 )
δ0 (x, y)) = −1, if (x, y) ∈ L(τ1 ) (34)
0, otherwise


159 Where G2 = {(ImedL , ImedG ), (mL , MG ), (a2k+1 , B2k+1 )}, k ∈ {0, 1, 2, 3} and τ2 , is a positive parameter
160 (threshold)
161

162 •Step 4 : We will proceed exactly as in the case of step 2 the binary codes are grouped in two separate
163 histograms as indicated by the equations ([Link].39) and ([Link].40) and, in the last step, the histograms are
164 concatenated to form the entity Vector mentioned in the equation ([Link].41), the codes of the first two upper
165 and lower descriptors called structural topology models ST P (U τ2 ) and ST P (U τ2 ) are given in the equations
166 ([Link].35)) and ([Link].36) follows:
3
X
ST PUτ 2 (xc , yc ) = δ3 (mL , MG )25 + δ3 (ImedL , ImedG )24 + 2k δ3 (a2k+1 , B2k+1 ) (35)
k=0
167
3
X
ST P Lτ 2 (xc , yc ) = δ4 (mL , MG )25 + δ4 (ImedL , ImedG )24 + 2k δ4 (a2k+1 , B2k+1 ) (36)
k=0

168 Where δ3 (., .) and δ4 (., .) are two binary thresholding functions in this case are:


1, if (x, y) ∈ U (τ2 )
δ3 (x, y)) = (37)
0, otherwise
169


1, if (x, y) ∈ L(τ2 )
δ4 (x, y)) = (38)
0, otherwise

170 Where U (τ2 ) and L(τ2 ) represent the two sets defined in the equations([Link].32)and ([Link].33)
171

9
172 The two descriptor histograms hST P (U τ2 ) and hST P (Lτ2 ) illustrated in equations([Link].39)and ([Link].40) are
173 finally concatenated to form the final hST P (τ2 ) future vector as illustrated in equation ([Link].(41). hST P (τ2 )
174 generates 2.26 possible different patterns
M
X −1 N
X −1
hST P (U τ2 ) (k2 ) = ϕ(ST PU τ 2 (i, j), k2 ) (39)
i=0 j=0
175
M
X −1 N
X −1
hST P (Lτ2 ) (k2 ) = ϕ(ST PLτ 2 (i, j), k2 ) (40)
i=0 j=0

hST P (τ2 ) = hST P (U τ2 ) ||hST P (Lτ2 ) (41)


1, if x = y
176 where: ϕ(x, y) = , k2 ∈ [0, 26 ]
0, otherwise
177

178 •Step 5 : The four descriptor histograms hST P (U τ1 ) , hST P (Lτ1 ) , hST P (U τ2 ) and hST P (Lτ2 ) are finally con-
179 catenated to form the final hST P (τ1 ,τ2 ) future vector as illustrated in Eq. (42).STP(τ1 , τ2 ) generates 2.27 +2.26
180 possible different patterns.

hST P (τ1 ,τ2 ) = hST P (U τ1 ) ||hST P (Lτ1 ) ||hST P (U τ2 ) ||hST P (Lτ2 ) (42)
181 We will also concatenate the four descriptors histograms mentioned above in double and triple form, as in the
182 following equations, so as to submit them to experimental tests to determine the best

hST P (U τ1 ,Lτ1 ) = hST P (U τ1 ) ||hST P (Lτ1 ) (43)

183

hST P (U τ1 ,U τ2 ) = hST P (U τ1 ) ||hST P (U τ2 ) (44)

184

hST P (U τ1 ,Lτ2 ) = hST P (U τ1 ) ||hST P (Lτ2 ) (45)

185

hST P (Lτ1 ,U τ2 ) = hST P (Lτ1 ) ||hST P (U τ2 ) (46)

186

hST P (Lτ1 ,Lτ2 ) = hST P (Lτ1 ) ||hST P (Lτ2 ) (47)

187

hST P (U τ2 ),Lτ2 ) = hST P (U τ2 ) ||hST P (Lτ2 ) (48)

hST P (U τ1 ,Lτ1 ,U (τ2 ) = hST P (U τ1 ) ||hST P (Lτ1 ) ||hST P (U τ2 ) (49)

188

hST P (U τ1 ,Lτ1 ,Lτ2 ) = hST P (U τ1 ) ||hST P (Lτ1 ) ||hST P (Lτ2 ) (50)

189

hST P (U τ1 ,U τ2 ,Lτ2 ) = hST P (U τ1 ) ||hST P (U τ2 ) ||hST P (Lτ2 ) (51)

190

hST P (Lτ1 ,U τ2 ,Lτ2 ) = hST P (Lτ1 ) ||hST P (U τ2 ) ||hST P (Lτ2 ) (52)

191 4 Experimental results and discussions

192 In this section, we compare our proposed descriptors with descriptors representing the dierent variants of LBP.
193 To this end, we conduct some experimental steps to prove the optimization of the parameters,improvement,
194 and the efficiency provided by our methods. The following subsections describe:
195 4.0.1 Texture datasets used in the experiments;
196 4.0.2 Distance measure and classification ;
197 4.0.3 Experiment protocol.
198 4.0.4 Optimization of parameters for the proposed methods;
199 4.0.5 Parameters optimization for parametric methods;
200 4.0.6 summary of texture descriptors tested and compared with our proposed descriptor;
201 4.0.7 The implementation, execution and reproducibility of the research;
202 5 Conclusion

10
203 4.0.1 Texture datasets used in the experiments
204 For evaluating the proposed method, twelve datasets are selected consisting of RSMAS ([Link]
205 [13],BarkTex[14],(Brodatz,Jerry Wu,KTH-TIPS,KTH-TIPS2b,OuTeX TC13 , VisTex) , ( USPTex , CUReT
206 , Kylberg ) and XU are the same as those used respectively in (fernandez et al)(2013)[15], El merabet et
207 al[2019][16] and Zhao et al[17],The main properties of each dataset are summarized in Table 2

No. Name Classes Samples Total samples Sample resolution Image format Predefined
per class (pixels) train/test sets?
. . . continued
1 Brodatz 13 16 208 256×256 Monochrome (TIFF) No
2 Jerry Wu 39 4 156 256×256 Color (BMP) No
3 Kylberg 28 160 4480 576×576 Monochrome (BMP) No
4 XUH 25 40 1000 1280×960 Monochrome (PNG) No
5 KTH-TIPS 10 4 40 100×100 Monochrome (PNG) No
6 KTH-TIPS2b 11 16 176 100×100 Color (PNG) No
7 CUReT 61 92 5612 200×200 Monochrome (PNG) No
8 USPtex1.0 191 12 2292 128×128 Color (PNG) No
9 NewBarkTex 06 68 1632 64×64 Color (BMP) No
10 Outex-TC13 68 20 1360 128×128 Color (BMP) No
11 VisTex 167 16 2672 128×128 Color (BMP) No
12 RSMAS 14 - 308 256×256 Color (PNG) No

Table 1: Summary of the image datasets considered in the experiments

208 4.0.2 Distance measure and classification

209 After the feature extraction step and to avoid unfair comparisons, in the classification process we use the nearest
210 neighbor method (1-NN) because of its efficiency and flexibility in experimental protocols. This clas-sification
211 strategy is particularly suitable for feature comparison purposes due to the absence of tuning pa-rameters, easi-
212 ness of implementation and other desirable asymptotic properties (Fernández et al., 2013). In our classification
213 problem, we will retain the class that has the minimum distance between two normalized frequency histograms.
214 In this context the distance L1 was used (cf. Eq. 39)
215

X
dL1 (, hj ) = |hi (k) − hj (k))| (53)
k

216 where hi ={h1i ,h2i ,...,hK 1 2 K

i } is the query feature vector and hj ={hj ,hj ,...,hj } the target feature vector.

217 4.0.3 Experiment protocol

218 To evaluate the classification performance of texture descriptors, we used a standard evaluation protocol for
219 each dataset (Fernández et al., 2013).
220 This base has been divided into two disjoint sub bases where 50% , of samples of each dataset (for each class)
221 were randomly selected for training and the remaining 50% for test-ing. The images of the validation sets are
222 then sorted according to the rule of the nearest neighbor (1 − N N ), with a distance of L1 the procedure is
223 repeated 100 times (100 different subdivisions into training and vali-dation sets) and the average accuracy is
224 reported as the final result for each dataset.

225 4.0.4 Optimization of parameters for the proposed methods

226 let D = {(x, y) ∈ N2 | (x,y) parameters of STP(x,y)}, let f , g D → R be two functions defined as :
1
P13 k
227 f (x, y) = G/C(x, y) and g(x, y)=M (x, y) = 13 k=0 CRD (x, y)

228 where G/C(x, y): is the Gain by Comparison obtained by the classification of wilkoxon,CRDk (x, y): indi-
1
P13 k
229 cates the classification rate of the k-th tested data set mentioned in the (Table.2) , and M (x, y) = 13 k=0 CRD (x, y)
k
230 is the average of CRD (x, y) for all k = 1, ..., 13

STP(x,y) 1 2 3 4 5 6 7 8 9 10 11 12 M(x,y) G /C(x, y)

continued . . .

11
 . . . continued
STP(x,y) 1 2 3 4 5 6 7 8 9 10 11 12 M(x,y) G /C(x, y)
STP(0,0) 99.82 99.54 99.73 95.21 95.64 93.63 94.24 89.03 85.39 84.21 74.84 71.38 90.22 24.22
STP(0,1) 99.87 99.59 99.72 95.46 95.78 93.82 94.30 90.25 86.02 85.29 75.62 72.53 90.69 37.24
STP(0,2) 100 99.65 99.74 95.95 95.84 94.83 94.47 91.23 87.09 85.72 76.12 73.82 91.21 52.86
STP(0,3) 100 99.81 99.73 96.28 95.67 94.69 94.30 91.85 87.05 84.96 76.18 73.81 91.19 56.00
STP(0,4) 100 99.85 99.77 96.46 95.37 94.83 94.12 91.99 86.44 84.74 76.04 73.55 91.10 55.00
STP(1,0) 99.98 99.56 99.75 95.67 95.62 94.5 94.62 90.11 86.86 84.34 76.29 71.86 90.76 42.00
STP(1,1) 99.94 99.63 99.73 95.80 95.57 95.08 94.57 90.75 87.33 84.82 76.79 72.81 91.07 51.56
STP(1,2) 100 99.85 99.75 96.11 95.67 95.35 94.55 91.43 87.88 84.94 77.24 74.23 91.42 66.15
STP(1,3) 100 99.85 99.76 96.22 95.59 95.28 94.33 91.98 87.43 84.56 76.91 74.48 91.37 64.84
STP(1,4) 100 99.85 99.78 96.55 95.22 95.03 94.03 91.98 87.00 84.07 76.57 73.38 91.12 56.25
STP(2,0) 100 99.90 99.75 96.00 95.17 94.77 94.76 90.71 86.04 84.11 76.43 70.72 90.70 52.86
STP(2,1) 100 99.90 99.74 96.01 95.26 95.56 94.61 90.99 86.56 84.73 76.87 72.08 91.03 57.55
STP(2,2) 100 99.90 99.73 96.11 95.19 95.94 94.52 91.36 86.75 84.64 76.96 72.97 91.17 60.68
STP(2,3) 100 99.90 99.77 96.20 94.87 95.99 94.22 91.52 86.62 84.40 76.60 72.72 91.07 59.11
STP(2.4) 100 99.90 99.75 96.00 95.17 94.77 94.76 90.71 86.04 84.11 76.43 70.72 90.70 43.49
STP(3,0) 100 99.90 99.79 95.90 94.82 94.65 94.74 90.51 85.33 83.12 76.12 71.40 90.52 39.84
STP(3,1) 100 99.90 99.76 95.95 94.74 95.30 94.58 90.64 86.07 84.03 76.54 72.39 90.83 48.70
STP(3,2) 100 99.90 99.73 96.09 94.60 95.72 94.31 90.82 85.85 84.29 76.58 72.71 90.88 48.18
STP(3,3) 100 99.90 99.78 96.19 94.33 95.45 94.03 91.26 85.97 83.30 76.02 71.81 90.67 45.05
STP(3,4) 100 99.90 99.80 96.15 94.18 95.11 93.61 90.82 85.56 82.91 75.69 70.09 90.32 37.24
STP(4,0) 100 99.90 99.82 95.68 94.53 94.86 94.54 90.16 85.09 82.98 75.97 70.45 90.33 33.85
STP(4,1) 100 99.90 99.79 95.62 94.67 94.67 94.41 90.47 86.03 83.59 76.40 71.64 90.60 39.32
STP(4,2) 100 99.90 99.77 95.72 94.42 94.93 94.13 90.42 85.69 83.49 76.62 71.85 90.58 36.20
STP(4,3) 100 99.90 99.82 95.74 94.16 95.18 93.74 90.59 85.29 82.70 75.93 71.03 90.34 34.63
STP(4,4) 100 99.76 99.81 95.88 93.82 94.92 93.24 90.01 85.44 81.80 75.40 69.34 89.95 28.12
STP(4,5) 100 99.76 99.85 95.86 93.52 95.18 92.93 89.35 85.24 81.46 75.14 68.31 89.72 27.86
STP(5,5) 100 99.90 99.82 95.92 93.62 95.12 93.06 90.29 85.85 83.45 75.05 69.31 90.12 32.03
STP(5,6) 100 99.81 99.86 95.59 92.75 94.50 92.22 88.18 84.47 79.73 73.93 66.81 88.99 21.61
STP(6,7) 100 99.81 99.87 95.27 92.22 94.51 91.32 86.52 83.32 77.05 73.16 65.14 88.18 19.79
STP(7,8) 100 99.81 99.88 95.08 91.42 93.40 90.43 85.30 82.11 74.34 72.35 63.22 87.28 17.45
STP(8,9) 100 99.31 99.87 95.01 90.62 93.06 89.56 84.39 81.02 71.37 71.36 60.89 86.37 12.50
STP(9,10) 100 99.17 99.86 95.08 89.88 93.01 88.81 83.21 79.87 68.39 70.42 58.72 85.54 10.42
STP(10,11) 100 98.87 99.87 94.84 89.35 93.15 87.88 81.95 78.93 65.81 69.63 57.79 84.84 07.81

Table 2: Classification Accuracy (%) has different parameter values (x, y) of the proposed method for all
databases /Classification accuracy obtained with TSP features for different parameter pair values (x, y)

231 the pair (x, y) represents the parametric values of our method determined by ourselves, then we try to
232 optimize the two functions f and g defined in (subsubsection 1.2.2) , here we have to ploting the two graphs
233 of f and g in order to determine the pair(x0 , y0 ) that maximize f and g in this case STP(x0 , y0 ) and (x0 , y0 ))
234 represent the best gain and the optimal point, respectively. Indeedbased on the experimental results of the tests
235 on the data sets shown in ( Table. 1), through which we obtained the optimal parameters, the values indicated
236 in light gray, we also plotted the graphs of the functions f and g, thus we obtained the optimal solution x =
237 1 and y = 2, This corresponds to f (1,2) = 91.42% and g (1,2) = 66.15% as shown in (Figure 6) and (Figure
238 7),Therefore, in all the remaining work we use the optimal parameters (1, 2).

239 4.0.5 Parameters optimization for parametric methods

240 Some methods summarized in Table 3 are methods that depend on one or two parameters. Therefore, the
241 composition of these user-defined parameters can have a significant impact on their results or performance, in
242 which case particular attention should be paid to these methods. For our methods, we have experimentally
243 identified the optimal parameters of τ1 ∈ [1, 10] and τ2 ∈ [2, 11] at values {1, 2},where (τ1 <τ2 ), see table. 2,
244 Figure .4 and Figure .5 above,

Descriptors Ref. Set of values Optimal values Descriptors Ref. Set of values Optimal values
LQP [12] τ2 ∈ [4, 6] {2, 6} CS-LMP [18] τ ∈ [1, 10] 2
LESTP [19] τ ∈ [1, 10] 1 LECTP [20] τ ∈ [1, 10] 1
AELTP [21] τ ∈ [1, 10] 3 CBP [22] τ ∈ [1, 10] 10
continued . . .

12
 . . . continued
Descriptors Ref. Set of values Optimal values Descriptors Ref. Set of values Optimal values
CS-LBP [23] τ ∈ [1, 10] 6 CS-TSτ [24] τ ∈ [1, 10] 9
ICS-TS [24] τ ∈ [1, 10] 8 LTP [10] τ ∈ [1, 10] 4
ILTP [11] τ ∈ [1, 10] 2 QBP [25] τ ∈ [1, 10] 2
GTUC [26] τ ∈ [1, 10] 9 dLBP [27] α ∈ {0, 45, 90, 135} 0
nLBPd [27] d ∈ [1, 4] 1 LQCH [17] x ∈ {2, 6}, y ∈ {3, 4} {6, 3}, {6, 4}

Table 3: Optimal parameter values of parametric descriptor on all tested datasets

245 4.0.6 summary of texture descriptors tested and compared with our proposed descriptor

Descriptors \Datasets Ref. 1 2 3 4 5 6 7 8 9 10 11 12

STP(1.2) this proposed 100.00 99.85 99.75 96.11 95.67 95.35 94.55 91.43 87.88 84.94 77.24 74.23
STPU 1+L1 +L2 this proposed 100.00 99.90 99.73 95.99 95.82 96.07 94.13 91.58 87.54 84.43 77.30 75.35
STPU 1+L1 +U2 this proposed 100.00 99.85 99.78 95.94 95.78 95.20 94.21 91.44 87.13 84.6 77.74 73.32
STPL1+U2 +L2 this proposed 100.00 99.68 99.68 95.78 95.14 94.88 94.23 90.54 86.95 84.53 75.60 73.08
STPU 1+U2 +L2 this proposed 100.00 99.69 99.61 95.88 95.30 94.73 94.14 90.23 86.54 85.17 75.85 72.58
STPU 1+L1 this proposed 100.00 99.68 99.75 95.53 95.77 95.82 92.76 91.15 84.65 83.22 77.49 74.38
STPL1+U2 this proposed 100.00 99.90 99.55 95.59 95.52 95.18 93.45 90.20 86.69 83.98 76.05 73.16
STPU 1+L2 this proposed 99.99 99.59 99.71 95.79 95.09 94.47 93.78 90.38 86.17 84.04 75.74 72.46
STPU 1+U2 this proposed 99.94 99.62 99.56 95.24 94.83 93.97 93.38 89.33 84.73 84.51 76.01 70.95
LCCSMP [28] 99.90 98.79 99.71 94.03 95.59 93.32 94.20 88.30 85.66 84.26 74.93 70.42
STPL1+L2 this proposed 100.00 99.28 99.61 95.29 94.43 95.58 92.61 89.30 85.25 81.86 74.07 70.68
STPU 2+L2 this proposed 100.00 99.04 99.39 95.05 94.08 93.89 93.25 87.82 85.57 83.93 73.09 70.01
ILTP [11] 100.00 98.79 99.74 95.34 94.44 93.91 91.92 89.37 81.63 80.34 74.64 69.12
LCxMSP [28] 99.78 98.90 99.72 93.83 95.10 91.15 93.34 86.21 84.78 83.61 74.01 71.02
CLBP-S-MxC [29] 99.94 98.86 99.73 94.08 94.35 94.05 91.63 87.56 82.96 81.12 73.29 68.17
MNTCDP [30] 99.93 99.37 99.52 92.92 95.15 90.93 89.70 85.29 81.33 81.14 75.815 74.88
LCvMSP [28] 99.88 98.74 99.38 93.32 94.17 92.83 92.3 86.55 84.60 83.02 73.92 68.79
STPU 1 this proposed 99.96 98.87 99.46 94.05 94.62 94.69 89.68 87.87 80.22 79.67 73.87 72.38
RALBGC [31] 99.75 97.63 99.46 93.35 95.06 93.25 92.51 87.08 85.04 82.55 75.51 68.83
STPL1 this proposed 99.87 99.36 99.19 94.10 95.06 92.91 89.66 87.05 79.34 81.99 75.42 68.31
ARCS-LBP [35] 99.93 98.83 99.60 93.22 94.44 93.28 91.25 84.49 80.20 82.44 71.31 68.04
STPU 2 this proposed 99.99 99.38 98.92 93.84 92.87 93.92 90.00 85.33 84.76 81.71 70.06 66.73
BSIF [33] 99.94 98.28 99.52 92.84 95.21 92.86 90.41 82.43 79.50 80.30 75.22 68.09
LTP [10] 99.99 98.13 99.60 94.52 93.79 92.92 91.59 87.38 79.54 78.53 72.91 65.81
IBGC1 [15] 99.79 97.94 99.04 92.51 94.04 93.1 91.15 84.64 82.02 78.78 74.59 70.36
STPL2 this proposed 99.93 97.73 98.96 93.69 92.55 92.70 91.33 83.94 82.86 82.43 71.69 67.55
SLBP [38] 99.83 98.29 99.05 92.13 92.96 91.36 90.87 84.39 80.82 80.03 72.23 66.60
LQP [12] 100.00 98.67 99.69 93.56 90.78 93.17 88.08 86.22 78.87 79.87 71.14 63.96
RLBGC [31] 99.90 97.18 98.97 92.55 93.50 92.77 90.65 84.54 82.80 80.58 73.11 66.67
LDTP(2) [36] 99.78 98.22 99.09 91.58 93.66 90.47 90.34 82.63 79.58 78.69 73.82 69.05
MSJLBP [34] 100.00 99.55 99.44 94.57 91.34 91.40 90.48 85.69 69.69 81.08 67.24 69.73
ALBGC [31] 99.69 97.54 98.75 92.25 94.14 92.00 89.60 84.59 81.74 81.09 74.13 66.32
ILBP [37] 99.82 96.71 99.52 93.13 93.19 92.09 88.2 83.41 75.93 79.74 72.33 68.80
ELGS [38] 99.80 97.28 98.54 91.67 92.76 88.03 91.96 83.02 82.17 80.81 71.23 67.18
AELTP [10] 100.00 99.08 99.11 93.29 92.11 88.49 90.14 82.82 77.33 79.30 64.66 62.97
AHP [39] 100.00 98.36 99.56 95.61 89.52 89.92 94.37 83.34 65.80 79.84 64.07 58.52
CRLBP-S [40] 99.62 97.82 99.23 92.68 92.79 89.74 89.23 82.84 77.32 79.25 70.10 63.18
LPQ [41] 99.92 98.01 98.47 91.78 93.46 87.16 84.57 78.31 75.36 76.95 73.01 70.74
RCS-LBP [35] 99.78 97.24 99.28 92.01 93.31 91.44 88.93 80.55 77.17 80.62 68.25 65.03
LBP [9] 99.63 97.99 98.65 91.13 91.84 89.67 89.84 81.16 79.78 78.15 71.51 65.88
ACS-LBP [32] 99.86 98.01 99.04 91.71 92.98 91.28 87.50 80.15 77.54 80.74 68.12 63.62
BGC1 [15] 99.70 97.27 98.06 90.15 91.75 90.38 90.36 80.89 81.25 78.04 72.78 66.55
nLBPd [27] 99.70 97.27 98.06 90.14 91.73 90.24 90.36 80.88 81.29 78.02 72.77 66.55
RILBP [42] 99.63 97.73 98.52 91.33 91.91 89.16 89.66 80.54 78.81 78.30 70.74 65.94
continued . . . c

13
 . . . continued
Descriptors \Datasets Ref. 1 2 3 4 5 6 7 8 9 10 11 12
maLBP [43] 99.66 98.19 98.86 92.01 91.96 88.20 89.00 81.28 78.62 77.91 70.17 62.35
mdLBP [43] 99.66 98.19 98.86 92.01 91.96 88.20 89.00 81.28 78.62 77.91 70.17 62.35
DCLBP [44] 99.75 98.06 98.99 91.33 90.52 89.00 88.69 81.19 79.31 77.66 69.19 61.01
FDLBP [48] 99.66 95.85 97.94 91.52 94.40 83.02 92.85 77.87 77.59 77.67 66.22 69.23
MBP [45] 99.86 96.60 98.32 92.00 92.06 89.00 88.49 81.90 78.22 77.28 70.51 62.21
RLBP [46] 99.62 97.63 98.34 90.69 91.51 88.05 88.86 80.33 79.11 78.20 70.33 65.34
MRELBP [47] 99.53 98.49 96.94 94.29 89.84 90.51 85.76 79.34 70.00 79.28 62.81 74.78
NI-RD-LBP [49] 99.56 97.41 98.93 90.96 88.78 86.95 86.84 79.14 79.42 79.71 69.11 59.38
LBPHF [52] 99.65 98.01 97.62 93.04 91.98 87.61 87.53 80.48 72.71 78.20 68.53 66.29
LECTP [20] 100.00 98.69 98.64 94.50 85.04 88.59 82.18 80.7 61.85 75.59 68.07 64.11
LNDP [50] 99.89 96.81 97.77 90.04 91.34 88.72 88.40 79.52 77.77 75.01 71.57 63.59
NI/RD/CI/LBP [49] 99.74 97.91 98.90 91.60 88.75 87.55 85.81 77.62 77.86 78.30 66.99 58.25
LFDMag [52] 99.82 99.87 96.86 88.60 93.58 85.81 93.40 74.06 64.30 73.09 63.57 67.79
LGS [53] 99.09 97.09 97.25 89.99 89.94 84.19 87.64 78.73 77.63 78.93 69.03 67.50
LESTP [19] 99.79 99.18 95.80 95.23 89.28 88.80 79.06 78.97 59.12 77.32 67.93 69.40
SLGS [56] 99.21 96.96 96.90 87.98 89.01 85.17 88.33 78.49 78.91 79.19 69.28 68.55
DSLGS [57] 99.21 96.96 96.90 87.98 89.01 85.17 88.33 78.49 78.91 79.19 69.28 68.55
3DLBP [54] 99.58 97.27 99.01 91.23 86.23 89.03 80.48 80.20 76.90 76.55 69.24 62.26
LOOP [58] 99.42 98.28 96.10 89.69 90.14 85.80 82.50 77.21 76.62 77.69 68.62 64.48
LMEBP [55] 99.50 99.08 95.76 94.28 89.46 87.81 77.64 78.11 58.77 76.28 66.98 70.16
SMEPOP [61] 99.12 97.63 98.21 90.16 89.99 86.08 86.30 76.27 76.83 76.08 66.66 61.47
LQPAT [59] 99.16 96.15 98.03 89.49 91.57 83.76 89.80 76.74 79.32 75.68 67.71 60.55
WLD [60] 99.26 95.46 97.92 88.71 82.63 90.98 75.38 81.00 75.28 84.53 68.71 60.54
BGC2 [15] 98.92 97.51 96.91 89.47 91.43 82.84 90.36 75.35 75.79 76.16 64.74 58.95
DRLBP [62] 97.90 97.27 96.68 93.22 92.18 89.41 86.06 78.14 60.78 76.96 66.22 65.46
MD-LBP [63] 99.68 97.13 98.11 91.26 89.98 82.00 86.12 73.02 74.30 75.35 62.18 62.00
RD-CI-LBP [51] 99.63 98.00 98.40 89.95 87.38 84.74 82.36 73.08 75.33 75.10 63.11 54.71
BGC3 [15] 99.23 96.78 97.14 89.42 90.00 84.72 87.20 74.16 75.22 75.01 63.28 57.38
LQCH [17] 99.09 97.33 95.76 92.17 88.10 90.51 83.75 75.27 63.19 75.84 60.14 62.68
AELBP [64] 99.64 97.4 97.77 88.96 89.00 79.64 82.86 69.21 71.06 74.43 58.08 61.11
CRLBP-M [43] 99.65 95.97 96.87 86.38 88.06 82.50 78.98 71.43 71.72 74.82 62.3 55.92
AECLBP-S [64] 99.66 97.28 97.58 88.63 88.66 78.63 81.62 68.58 70.22 74.04 57.17 60.39
CS-TSdelta [24] 98.04 93.74 97.21 88.96 85.07 86.76 80.07 74.23 76.40 72.51 63.73 55.36
CSQP [65] 98.56 96.74 97.53 88.33 87.54 80.13 85.10 69.90 74.22 72.34 61.75 57.29
STS [66] 99.50 94.94 96.86 89.60 87.69 82.15 82.06 71.37 74.92 73.95 62.87 52.27
NI-LBP [51] 98.85 96.04 97.59 88.80 84.48 81.75 75.52 65.85 69.77 76.50 60.51 52.14
LDTP [67] 97.83 94.76 98 87.44 83.31 87.70 82.88 66.50 72.76 68.8 57.94 50.19
CELDP [69] 98.15 89.19 97.95 88.02 82.13 85.73 62.71 71.92 72.77 77.80 60.94 50.57
CCR [78] 98.40 93.40 98.43 86.02 86.68 84.42 90.18 63.92 69.77 69.29 51.05 44.73
DCP [70] 99.46 97.15 93.49 84.10 88.77 75.90 83.81 64.92 69.79 67.31 54.40 57.90
NI-CI-LBP [51] 98.96 97.31 97.47 88.68 82.98 82.07 74.80 63.79 67.76 74.17 57.91 48.88
AECLBP-M [64] 98.59 97.36 95.58 83.11 86.11 81.01 76.12 68.40 67.81 72.43 56.06 55.97
LDP [71] 98.87 96.47 93.03 84.29 84.90 85.27 77.54 64.45 68.16 69.30 66.85 51.08
CBP [72] 97.86 94.42 97.48 86.34 79.37 86.74 70.59 70.17 72.42 66.45 58.19 50.27
GTUC [76] 98.27 95.38 96.33 87.18 79.83 86.47 65.52 68.40 71.71 59.95 61.23 52.96
LDN [75] 98.12 93.92 94.75 84.80 85.89 80.87 77.86 63.84 73.29 72.93 56.63 52.16
LFDPhase [73] 94.18 97.91 94.42 80.78 89.38 76.64 94.31 53.17 57.86 63.51 47.74 59.95
MMEPOP [80] 97.00 93.94 92.54 84.87 84.41 81.72 72.90 67.29 67.48 70.10 63.64 55.22
GLBP [76] 98.88 93.96 94.02 85.51 85.38 80.57 74.18 69.37 65.01 71.39 64.48 53.06
ELDP [78] 99.06 92.81 96.41 88.01 88.55 79.14 77.43 63.30 66.44 72.28 55.46 50.06
LDRP [77] 98.43 84.81 93.39 85.54 88.47 78.70 82.19 61.32 66.69 68.81 54.01 61.22
LCCP [79] 98.27 90.55 95.97 85.32 86.83 82.51 63.71 65.81 64.99 64.76 57.44 56.31
D-LBP [81] 98.21 95.65 95.71 82.30 75.68 82.65 66.03 62.56 66.88 68.86 62.70 54.31
XCSLBP [82] 98.12 91.62 97.54 84.51 69.88 85.41 69.56 60.29 68.40 54.14 60.81 47.96
QBP [25] 98.10 92.60 95.67 85.64 74.81 87.85 65.96 63.39 71.24 63.24 54.15 46.71
DC [83] 97.35 96.46 91.14 85.91 83.49 79.78 68.72 59.05 57.50 68.84 54.34 52.97
LDGP [84] 94.60 91.86 94.09 84.01 83.24 77.39 74.47 59.98 68.66 70.09 54.40 47.96
LWP [86] 98.90 60.14 98.88 82.90 77.10 79.94 60.69 53.05 54.22 52.72 51.04 46.29
continued . . . c

14
 . . . continued
Descriptors \Datasets Ref. 1 2 3 4 5 6 7 8 9 10 11 12
LDEP [87] 98.9 60.14 98.88 82.90 77.10 79.94 60.69 53.05 54.22 52.72 51.04 46.29
MTS [85] 95.44 90.92 94.32 83.29 82.28 75.30 75.29 60.46 69.62 69.12 53.54 42.73
GLTCS+ [88] 95.18 91.94 89.69 83.02 81.06 72.24 66.65 58.52 69.56 70.82 52.83 46.51
dLBPa [27] 91.16 90.21 93.00 74.73 83.04 78.69 85.65 54.87 49.62 67.88 46.76 53.62
ELDPd [90] 96.63 88.12 94.08 83.24 84.53 73.72 73.08 56.74 66.44 65.32 47.60 44.32
LDNP [91] 93.51 85.09 89.40 80.65 80.58 77.75 63.99 59.16 64.54 69.47 51.43 48.20
OTF [92] 94.12 95.60 89.86 79.56 79.14 73.60 68.84 51.41 52.77 57.79 43.17 57.92
RT [93] 94.62 93.68 85.73 78.59 74.96 77.11 60.24 55.66 56.85 69.37 50.44 54.31
ID-LBP [95] 96.95 91.67 91.88 79.13 79.27 75.47 61.22 54.39 61.14 66.91 53.14 46.02
LBDP [94] 95.74 57.6 90.78 84.56 70.73 80.42 61.25 42.72 50.23 72.00 53.35 50.27
BTCS+ [95] 96.20 88.09 94.98 74.96 77.07 78.06 74.40 49.45 64.46 64.08 40.46 38.38
ICS-TSdelta [24] 96.00 85.95 90.86 78.14 71.17 79.11 55.07 52.37 68.07 66.97 44.77 44.81
LBC [96] 93.66 91.86 83.27 77.37 72.14 72.09 55.89 52.81 56.17 68.12 47.72 52.03
LDENP [97] 93.26 88.74 87.77 69.95 72.31 65.80 52.97 41.24 61.16 62.77 41.86 44.29
MLOP [20] 83.78 63.38 89.45 76.07 59.97 74.31 48.44 42.78 46.02 51.92 40.76 39.9
LOSIB [98] 89.59 56.38 86.29 63.56 52.28 69.25 37.47 36.50 58.35 48.37 30.84 26.21
FLBP [99] 66.55 42.83 45.40 52.68 40.29 58.15 21.91 25.41 37.77 38.00 22.70 31.79
CS-LBP [23] 58.57 17.62 37.53 24.48 28.75 39.55 14.33 8.57 38.50 19.45 07.54 15.38
VAR-LBP [100] 43.29 11.67 30.01 33.32 24.31 45.64 14.67 14.00 38.48 33.72 12.56 26.51
LBPD [101] 47.16 43.45 33.01 20.02 31.28 27.32 11.93 9.37 32.29 16.92 7.91 19.68
CS-LMP [18] 49.47 16.29 29.88 23.73 22.61 44.41 10.15 06.79 32.64 17.94 05.81 17.66
CI-LBP [19] 39.95 10.92 24.03 15.37 25.49 28.52 06.81 02.81 23.68 07.49 03.89 12.95
AECLBP-C [102] 40.24 10.28 22.52 14.70 25.38 27.55 06.14 03.02 23.97 07.45 03.22 14.01
RCRLBP-C [103] 33.62 14.63 21.41 12.63 26.94 20.23 06.41 02.81 22.36 05.86 03.29 10.84

Table 4: Ranking results for each data set tested. The methods that are at the top of the rankings are
highlighted in light grey

246 Our objective is to study the stability of the performance of proposed methods by evaluating them on several
247 sets of data in Table I. Table IV summarizes the results of the classification of the total descriptors tested in the
248 current work, while the experimental results are summarized in Table V, which reflects the average accuracy of
249 each method evaluated for each set of data tested. In the figure (x) We present the results of the classification
250 in each data set according to the average accuracy in the form of graphs. So that the results can be drawn as
251 follows:
252 • At the beginning of our observations and on the basis of the experimental results of the two above mentioned
253 tables, we note that CRLBP-C, CI-LBP, AECLBP-C, VAR-LBP, LBPD, CS-LMP, CS-LBP, FLBP, MLOP and
254 LOSIB. Many other descriptors also perform poorly in all texture test datasts and therefore tend to be in the
255 lowest order.
256 • It also appears in both Tables IV and V and, with the exception of the previously mentioned descriptors,
257 where the corresponding average accuracy in each data set is less than 90%, all remaining descriptors tested
258 have very promising results for the first time. Brodatz dataset (dataset 1 in Table IV) for the same Brodatz
259 dataset some parametric and non-parametric descriptors such as LTP, LECTP, AHP, AELTP, MSJLBP, LQP
260 and ILTP, including the first nine proposed dissectors have also achieved average accuracy of 100%, and we also
261 find that the first nine proposed descriptors as well as the two descriptors LFDMag and MSJLBP for the Jerry
262 Wu data set (dataset 2 in Table IV), in the same context the first eight descriptors as well as the LTP, BSIF,
263 AHP, ARCS-LBP, LQP, LCCSMP, CLBP/S/MxC and ILTP descriptors for the Kylberg dataset (dataset 3 in
264 Table IV) as well as the proposed first eleven descriptors and the LESTP, ILTP and AHP descriptors for the
265 XUH dataset (dataset 4 in Table IV), the first nine proposed descriptors, and the BSIF, MNTCDP, RALBGC
266 and LCCSM descriptors for KTH-TIPS dataset (dataset 5 in Table IV) and the first six proposed KTH-TIPS2b
267 datasets (dataset 6 in Table IV) all exceeded an average accuracy of 95.5% . In the other six remaining data
268 sets, it can be concluded that the first five proposed descriptors and the AHP, LFDMag and LCCSM descriptors
269 for the Curetgrey dataset (dataset 7 in Table IV) exceed an accuracy of 94%, we can also see that the average
270 accuracy of the first eight proposed descriptors for the USPtex1.0 dataset (dataset 8 in Table IV) exceed an
271 average accuracy of 90%, in this way, we see the first nine proposed descriptors and the two LCCS and RQLBGC
272 descriptors for the NewBarkTex dataset (dataset 9 in Table IV) exceed an accuracy of 85%, whereas the first
273 ten descriptors proposed, and the two descriptors LCCSM and WLD, for the Outex-TC-00013 dataset (dataset
274 10 in Table IV) exceed an average accuracy of 83%, apparently we can signaled that the first ten descriptors
275 proposed and the descriptors MNTCDP, RALBGC, and BSIF for the VisTex dataset (dataset 11 in Table
276 IV) exceed an average accuracy of 75% and finally the first six descriptors proposed and the two descriptors

15
 f(1,2)=91,42
92,000 70
g(x,y)=66,15
90,000 60

g(x,y)= G/C(x,y)
50
f(x,y)=M(x,y)

88,000
40
86,000
30
84,000
20
82,000 Optimal value
10 Optimal value
80,000 0

STP(10,11)
STP(9,10)
STP(0,0)
STP(0,1)
STP(0,2)
STP(0,3)
STP(0,4)
STP(1,0)
STP(1,1)
STP(1,2)
STP(1,3)
STP(1,4)
STP(2,0)
STP(2,1)
STP(2,2)
STP(2,3)
STP(2,4)
STP(3,0)
STP(3,1)
STP(3,2)
STP(3,3)
STP(3,4)
STP(4,0)
STP(4,1)
STP(4,2)
STP(4,3)
STP(4,4)
STP(4,5)
STP(5,5)
STP(5,6)
STP(6,7)
STP(7,8)
STP(8,9)

STP(0,0)
STP(0,1)
STP(0,2)
STP(0,3)
STP(0,4)
STP(1,0)
STP(1,1)
STP(1,2)
STP(1,3)
STP(1,4)
STP(2,0)
STP(2,1)
STP(2,2)
STP(2,3)
STP(2,4)
STP(3,0)
STP(3,1)
STP(3,2)
STP(3,3)
STP(3,4)
STP(4,0)
STP(4,1)
STP(4,2)
STP(4,3)
STP(4,4)
STP(4,5)
STP(5,5)
STP(5,6)
STP(6,7)
STP(7,8)
STP(8,9)
STP(9,10)
STP(10,11)
Descriptors Descriptors

(1) (2)

Figure 7: (1):Graphical optimization of the function f ,(2):Graphical optimization of the function g

Figure 8: Calculation of the STP with eight neighboring pixels )

277 MNTCDP and MRELBP for the RSMAS data set (dataset 12 in Table IV) exceed an average accuracy of 73%
278 • One of the most important observations we can point out in this paper is that the proposed method is
279 a parametric method based on two optimized parameters on the one hand and on the other hand on the
280 concatenation of four descriptors two to two disjoint from the topological and structural points of view, so that
281 we obtain a total of fifteen proposed descriptors, the first ten of which are more effective ,in terms of average
282 accuracy, dimensions and time of execution , which are mentioned in the results of Tables 4 and 5,and also in
283 the figures (1, ..., 12). which shows that our top ten descriptors (see table V) are the best among to other the
284 descriptors most popular and the most recent tested.

No Texture descriptor Ref. Year Acronym V/C Dim T.(mn)

1 Structural Topology Patterns this proposede 2019 STP(1,2) 0.96933 380 4.1
continued . . .

16
. . . continued
No Texture descriptor Ref. Year Acronym V/C Dim. T.(mn)
2 Structural Topology Patterns this proposede 2019 STPU1 L1 L2 0.96400 317 6.5
3 Structural Topology Patterns this proposede 2019 STPU1 L1 U2 0.96133 317 6.6
4 Structural Topology Patterns this proposede 2019 STPL1 U2 L2 0.93200 253 4.5
5 Structural Topology Patterns this proposede 2019 STPU1 U2 L2 0.93133 253 3.6
6 Structural Topology Patterns this proposede 2019 STPU1 L1 0.92800 254 3.8
7 Structural Topology Patterns this proposede 2019 STPL1 U2 0.92867 190 3.6
8 Structural Topology Patterns this proposede 2019 STPU1 L2 0.91800 190 6.0
9 Structural Topology Patterns this proposede 2019 STPU1 U2 0.89067 190 6.4
10 Structural Topology Patterns this proposede 2019 STPL1 L2 0.88267 190 7.0
11 Local Concave-and-Convex Micro-structures Pattern [28] 2018 LCCSMP 0.87600 2046 6.3
12 Structural Topology Patterns this proposede 2019 STPU2 L2 0.86133 126 4.6
13 Improved Local Ternary Patterns [11] 2010 ILTP 0.85000 1024 4.9
14 Local Concave-and-Convex Micro-structures Pattern [28] 2018 LCxMSP 0.83067 1023 4.0
15 Completed local binary patterns (S+MxC) [29] 2010 CLBP-S-MxC 0.82600 768 4.3
16 Mixed neighborhood topology cross decoded patterns [30] 2018 MNTCDP 0.81467 2048 4.6
17 Structural Topology Patterns this proposede 2019 STPU1 0.81133 127 4.0
18 Local Concave Micro-Structures Pattern [28] 2018 LCvMSP 0.80867 1023 3.3
19 Structural Topology Patterns this proposede 2019 STPL1 0.79333 127 5.7
20 Repulsive-and-repulsive local binary gradient contours [31] 2018 RALBGC 0.79000 1022 3.4
21 Attractive-and-repulsive center-symmetric LBP [32] 2019 ARCS-LBP 0.78667 126 5.1
22 Structural Topology Patterns this proposede 2019 STPU2 0.77600 63 4.5
23 Binarized Statistical Image Features [33] 2012 BSIF 0.76867 256 4.7
24 Local Ternary Patterns [10] 2007 LTP 0.76667 512 4.0
25 Improved binary gradient contours (1) [15] 2012 IBGC1 0.75400 510 3.9
26 Structural Topology Patterns this proposede 2019 STPL2 0.75267 63 4.5
31 Multi-scale Joint Encoding of Local Binary Patterns [34] 2013 MSJLBP 0.73667 3540 6.3
28 Local Quinary Patterns [12] 2010 LQP 0.72800 1024 4.4
29 Repulsive local binary gradient contours [31] 2018 RLBGC 0.72667 511 5.1
27 Soft histograms for LBP [35] 2007 SLBP 0.72067 256 4.7
30 Local directional ternary pattern [36] 2018 LDTP(2) 0.70733 1022 5.8
32 Attractive local binary gradient contours [31] 2018 ALBGC 0.69800 511 5.2
33 Improved LBP [37] 2013 ILBP 0.67800 511 5.4
34 Extended Local Graph Structure [38] 2016 ELGS 0.67400 512 5.4
35 Adjacent Evaluation LTP [10] 2015 AELTP 0.67267 512 5.3
36 Adaptive Hybrid Pattern [39] 2015 AHP 0.66933 120 9.3
37 Completed robust local binary pattern (Sign) [40] 2013 CRLBP-S 0.64200 256 6.0
39 Repulsive center-symmetric local binary patterns [32] 2019 RCS-LBP 0.64067 63 4.9
41 Attractive center-symmetric local binary patterns [32] 2019 ACS-LBP 0.63267 63 4.9
40 Local Binary Patterns [9] 1996 LBP 0.63133 256 4.1
42 Binary Gradient Contours (1) [15] 2011 BGC1 0.61800 255 4.9
43 Local Binary Patterns by Neighborhoods [27] 2015 nLBPd 0.61733 256 3.4
38 Local Phase Quantization [41] 2008 LPQ 0.61667 256 3.8
44 Rotation Invariant Local Binary Pattern [42] 2008 RILBP 0.60600 256 5.2
45 Multichannel Adder Based Local Binary Pattern [43] 2017 maLBP 0.60400 1024 8.7
46 Multichannel Decoder Based Local Binary Pattern [43] 2017 mdLBP 0.60400 2048 8.6
47 Multi-Scale Densely Sampled Complete LBP [44] 2013 DCLBP 0.59667 118 4.9
48 Median Binary Patterns [45] 2007 MBP 0.59067 511 10.1
49 Robust Local Binary Pattern [46] 2013 RLBP 0.58267 256 4.4
50 Median Robust Extended Local Binary Pattern [47] 2016 MRELBP 0.58133 200 5.1
51 Frequency Decoded Local Binary Pattern [48] 2017 FDLBP 0.56667 4096 6.9
52 LBP histogram Fourier [49] 2012 LBPHF 0.56667 76 6.1
53 Local Extreme Complete Trio Pattern [20] 2015 LECTP 0.55667 768 4.2
54 Local Neighborhood Difference Pattern [50] 2017 LNDP 0.55533 256 5.9
55 NILBP ∪RDLBP [51] 2012 NI-RD-LBP 0.54733 118 4.3
56 Local Frequency Descriptor Mag [52] 2013 LFDMag 0.54667 276 10.6
57 Local Extreme Sign Trio Pattern [19] 2015 LESTP 0.53933 512 15.8
58 NILBP ∪RDLBP ∪ CILBP [51] 2012 NI-RD-CI-LBP 0.53733 120 3.6
59 Local Graph Structure [53] 2013 LGS 0.51867 256 5.2
continued . . .

17
. . . continued
No Texture descriptor Ref. Year Acronym V/C Dim. T.(mn)
60 3D Local Binary Patterns [54] 2006 3DLBP 0.51533 1024 3.6
61 Local Maximum Edge Binary Patterns [55] 2012 LMEBP 0.51467 256 3.4
62 Symmetric Local Graph Structure [56] 2015 SLGS 0.50800 256 5.3
63 Difference Symmetric Local Graph Structure [57] 2015 DSLGS 0.50800 256 15.2
64 Local Optimal-Oriented Pattern [58] 2018 LOOP 0.50067 256 4.1
65 Local quadruple pattern [59] 2017 LQPAT 0.49400 512 4.7
66 Weber Local Descriptor [60] 2010 WLD 0.49267 617 4.6
67 Sign Maximum Edge Position Octal Pattern [61] 2015 SMEPOP 0.49067 501 12.6
68 Dominant Rotated Local Binary Patterns [62] 2016 DRLBP 0.48200 256 3.7
69 Binary Gradient Contours (2) [15] 2011 BGC2 0.46733 225 9.0
70 Multi-Direction Local Binary Pattern [63] 2019 MD-LBP 0.45733 255 6.1
71 RDLBP ∪ Center Intensity based LBP (CILBP) [51] 2012 RD-CI-LBP 0.45000 61 6.0
72 Binary Gradient Contours (3) [15] 2011 BGC3 0.43133 255 3.7
73 Local Quantization Code Histogram [17] 2016 LQCH 0.42600 10000 9.3
74 Adjacent evaluation of local binary pattern [64] 2015 AELBP 0.40333 256 6.4
75 Adjacent Evaluation Completed LBP (Sign) [64] 2015 AECLBP-S 0.37733 256 3.6
76 Center-Symmetric Texture Spectrum [24] 2011 CS-TS4 0.37400 81 10.1
77 Simplified Texture Spectrum [66] 2003 STS 0.37400 81 3.5
78 Center Symmetric Quadruple Pattern [65] 2017 CSQP 0.37267 256 4.4
79 Completed robust local binary pattern (Magnitude) [43] 2013 CRLBP-M 0.37067 256 3.8
80 Neighborhood Intensity based [51] 2012 NI-LBP 0.33800 59 6.7
81 Local Directional Texture Pattern [67] 2015 LDTP 0.33200 256 5.2
82 Coordinated Clusters Representation [68] 1996 CCR 0.32800 512 4.0
83 Complete Eight Local Directional Patterns [69] 2016 CELDP 0.32667 256 5.0
84 NILBP ∪ CILBP [51] 2012 NI-CI-LBP 0.31667 61 6.8
85 Dual-Cross Patterns [70] 2016 DCP 0.30533 512 5.2
86 Local Directional Pattern [71] 2010 LDP 0.30133 128 10.1
87 Adjacent Evaluation Completed LBP (Magnitude) [64] 2015 AECLBP-M 0.30067 256 7.0
*** 88 Centralized Binary Patterns [72] 2008 CBP 0.29800 32 3.5
89 Local Frequency Descriptors (Phase) [73] 2013 LFDPhase 0.29667 264 6.1
** 90 Gradient texture unit coding [74] 2004 GTUC 0.29133 4374 5.4
91 Local Directional Number Pattern [75] 2013 LDN 0.29067 256 54.3
92 Gradient Local Binary Patterns [76] 2011 GLBP 0.29067 256 4.7
93 Eight Local Directional Patterns [78] 2015 ELDP 0.28667 256 3.7
94 Magnitude Maximum Edge Position Octal Pattern [80] 2015 MMEPOP 0.27533 501 5.0
95 Local Directional Relation Pattern [77] 2017 LDRP 0.26267 1024 13.7
96 Local Convex-and-Concave Pattern [79] 2016 LCCP 0.25933 16 4.6
97 Direction Local Binary Patterns (D) [81] 2009 D-LBP 0.25133 16 4.6
98 Quad Binary Pattern [25] 2016 QBP 0.24133 16 11.2
* 99 eXtended Center-Symmetric LBP [82] 2015 XCSLBP 0.24000 16 4.3
100 Directional Coding [83] 2018 DC 0.22333 225 8.7
101 Local Directional Gradient Pattern [84] 2015 LDGP 0.21400 63 4.6
102 Modified Texture Spectrum [85] 2003 MTS 0.20067 16 4.3
103 Local Wavelet Pattern [86] 2015 LWP 0.18933 256 4.3
104 Local Diagonal Extrema Pattern [87] 2015 LDEP 0.18933 256 4.4
105 Gray Level Texture Co-occurrence Spectrum [88] 1996 GLTCS+ 0.18533 24 3.7
106 Directional Local Binary Patterns [27] 2015 dLBPa 0.18200 256 4.3
107 Enhanced Local Directional Patterns Dominant [90] 2013 ELDPd 0.17467 62 6.3
108 Local Directional Number Pattern [89] 2012 LDNP 0.16200 63 4.6
109 Oriented Template Features [91] 2012 OTF 0.15800 290 62.6
110 Rank transform [92] 1994 RT 0.15733 9 4.5
111 Local Bit-plane Decoded Pattern [93] 2016 LBDP 0.15733 256 4.1
112 Improved Direction Local Binary Patterns (ID) [94] 2009 ID-LBP 0.15667 16 7.5
113 Binary Texture Co-occurrence Spectrum [95] 1991 BTCS+ 0.14733 16 7.5
114 Improved Center-Symmetric Texture Spectrum [24] 2012 ICS-TSdelta 0.13867 18 3.6
115 Local Binary Count [96] 2012 LBC 0.12667 8 4.7
116 Local Diagonal Extrema Number Pattern [97] 2017 LDENP 0.09800 15 26.1
117 Magnitude Local Operator [20] 2016 MLOP 0.08067 256 4.7
continued . . .

18
 . . . continued
No Texture descriptor Ref. Year Acronym V/C Dim. T.(mn)
118 Local Oriented Statistics Information Booster [98] 2014 LOSIB 0.07200 8 4.2
119 Fuzzy Local Binary Patterns [99] 2013 FLBP 0.05467 256 6.3
120 Center-Symmetric Local Binary Patterns [23] 2006 CS-LBP 0.03867 16 5.6
121 LBP Variants [100] 2010 VAR-BP 0.03800 10 18.6
122 LBP Difference [101] 2014 LBPD 0.03333 256 11.5
123 Center Symmetric Local Multilevel Pattern [18] 2016 CS-LMP 0.02733 4 5.0
124 Center Intensity based LBP [19] 2012 CI-LBP 0.01200 2 13.1
125 Adjacent Evaluation Completed LBP (Center) [102] 2015 AECLBP-C 0.00800 256 12.6
126 Completed robust local binary pattern (Center) [103] 2013 CRLBP-C 0.00533 256 6.0

Table 5: Ranking results according to the normalized number of victories obtained by each tested method on
all the tested datasets using Wilcoxon based ranking test..

285 •Il It is interesting to note that the concatenation of the four descriptors proposed which gives a descriptor
286 is called ST PU1 + ST PL1 + ST PU2 + ST PL2 = STP(1, 2) is more efficient than the triple concatenation of the
287 descriptors which leads to four descriptors, namely ST PU1 + ST PL1 + ST PL2 = STPU1 L1 L2 , ST PU1 + ST PL1 +
288 ST PU2 = STPU1 L1 U2 , ST PL1 + ST PU2 + ST PL2 = STPU1 U2 L2 and ST PU1 + ST PU2 + ST PL2 = STPU1 U2 L2
289 , and these are better that the double concatenation of the proposed descriptors which gives six descrip-
290 tors which are ST PU1 + ST PL1 = STPU1 L1 , ST PL1 + ST PU2 = STPL1 U2 , ST PU1 + ST PL2 = STPU1 L2 ,
291 ST PU1 + ST PU2 = STPU1 U2 , ST PL1 + ST PL2 = STPL1 L2 and ST PU2 + ST PL2 = STPU2 L2 and finally
292 the proposed descriptors STPU1 ,STPL1 ,STPU2 and STPL2 so that they also obtained important results
293 whether it is a multilevel discretization or a pure binarization . as ILTP (13th), LTP (24th) and LQP (28th)
294 or CLBP-S-MxC (15th), MSJLBP (27th) and MRELBP (50th) (see table 5)
295

296

297 4.0.7 Implementation, execution and research reproducibility

298 Texture descriptors have been implemented in MATLAB, R2018b. The classification experiments were carried
299 out on a PC equipped with the Intel® Core TM i5-7200U CPU @ 2.50 GHz processor, 2701 MHz, 2 core
300 (s), 4 processor, RAM 12 GB. The execution of the scripts required more than (x) hours of computation time.
301 Fig.10:(c) illustrates the processing time (in minutes), including the feature extraction time, distance calcula-
302 tion, and 1-NN classification, taken by the first 50 ranked-ordered descriptors as well as the three-valued deep
303 characteristics over 16939 images making up the twelve datasets used
304

305 5 Conclusion
306 The main objective of this article was to propose a new approach for the classification of texture images. In this
307 respect, we have introduced a two-parameter topological structure with upper and lower directions. On the basis
308 of these methods, the operators proposed in this article ST PU1 , ST PL1 , ST PU2 and ST PL2 , constitute a family
309 of four descriptors, eleven of which were obtained after concatenation in double, triple and quadratic form. The
310 experiments and analyses we have carried out on these methods show their great advantages, from the point of
311 view of competitiveness and stability, thus making it possible to obtain the best results in terms of classification,
312 size and processing time, compared to other advanced methods, whether or not they are parameterised.

313 References
314 [1] Q. Ji, J. Engel, E. Craine, Texture analysis for classification of cervix lesions, Medical Imaging, IEEE
315 Transactions on 19 (11) (2000) 1144–1149.

316 [2] H. Anys, D. He, Evaluation of textural and multipolarization radar features for crop classification, Geoscience
317 and Remote Sensing, IEEE Transactions on 33 (5) (1995) 1170–1181
318 [3] W. Zhao, R. Chellappa, P. Phillips, and A. Rosenfeld. Face recognition: A literature survey. ACM Comput.
319 Surv., 35(4):399–458, 2003.

19
 Classification rate(%) Classification rate (%) Classification rate(%) Classification rate(%)

0
10
20
30
40
50
60
70
80
90
0
20
40
60
80
100
120

0
10
20
30
40
50
60
70
80
90
100
0
20
40
60
80
100
120
STP(1,2)
STP(1,2) STP(1,2)
STP-UL(1) STP(1,2)
STP-UL(1) STP-UL(1)
STP-L(1)-L(2) STP-L(1)-L(2)
STP-L(1)U(2)
STP-L(1)-L(2)
MNTCDP MNTCDP ILTP
MNTCDP
ARCS-LBP ARCS-LBP RALBGC
ARCS-LBP
STP-L(2) STP-L(2)
STP-L(2) IBGC1
MSJLBP MSJLBP MSJLBP
AHP
MSJLBP
AHP AHP
ACS-LBP CRLBP-S
ACS-LBP ACS-LBP
mdLBP mdLBP mdLBP nLBPd
MRELBP MRELBP MRELBP MBP
NI/RD/CI/LBP NI/RD/CI/LBP NI/RD/CI/LBP LNDP

XUH

Curet
DSLGS DSLGS DSLGS

(7)
(4)
(1)
DSLGS
Brodatz

(10)
LQPAT LQPAT LQPAT

Outex_TC13
WLD

Descriptors
RD-CI-LBP RD-CI-LBP

Descriptors
RD-CI-LBP

Descriptors
Descriptors
AECLBP-S AECLBP-S AECLBP-S
LQCH
LDTP LDTP LDTP STS
AECLBP-M AECLBP-M AECLBP-M NI-CI-LBP
LFDPhase LFDPhase LFDPhase LFDPhase
LCCP LCCP LCCP
LDGP
D-LBP
LDGP LDGP
dLBPa dLBPa LDEP
dLBPa
ID-LBP ID-LBP ID-LBP OTF
LDENP LDENP LDENP LBC
VAR-LBP VAR-LBP VAR-LBP VAR-LBP
CRLBP-C
CRLBP-C CRLBP-C

Classification rate(%) Classification rate(%) Classification rate (%) Classification rate(%)

0
10
20
30
40
50
60
70
80
90
0
10
20
30
40
50
60
70
80
90
100
0
20
40
60
80
100
120
0
20
40
60
80
100
120

STP(1,2)
STP(1,2) STP(1,2) STP(1,2)
STP-UL(1) STP-UL(1) STP-UL(1) STP-UL(1)
STP-L(1)-L(2) STP-L(1)-L(2) STP-L(1)-L(2) STP-L(1)-L(2)
MNTCDP MNTCDP MNTCDP MNTCDP
ARCS-LBP ARCS-LBP ARCS-LBP ARCS-LBP
STP-L(2) STP-L(2) STP-L(2) STP-L(2)
MSJLBP MSJLBP MSJLBP MSJLBP
AHP AHP AHP AHP
ACS-LBP ACS-LBP ACS-LBP ACS-LBP
mdLBP mdLBP mdLBP mdLBP
MRELBP MRELBP MRELBP MRELBP
NI/RD/CI/LBP NI/RD/CI/LBP NI/RD/CI/LBP NI/RD/CI/LBP
DSLGS DSLGS DSLGS DSLGS

VisTex
(8)
(5)
(2)
Jerry wu

KTH-TIPS

(11)
LQPAT

USPtex1.0
LQPAT LQPAT LQPAT

Descriptors
RD-CI-LBP
Descriptors

Descriptors
Descriptors
RD-CI-LBP RD-CI-LBP RD-CI-LBP
AECLBP-S AECLBP-S AECLBP-S AECLBP-S
LDTP LDTP LDTP LDTP
AECLBP-M AECLBP-M AECLBP-M AECLBP-M

20
LFDPhase LFDPhase LFDPhase LFDPhase
LCCP LCCP LCCP LCCP
LDGP LDGP LDGP LDGP
dLBPa dLBPa dLBPa dLBPa
ID-LBP ID-LBP ID-LBP ID-LBP
LDENP LDENP LDENP LDENP
VAR-LBP VAR-LBP VAR-LBP VAR-LBP
CRLBP-C CRLBP-C CRLBP-C CRLBP-C

Classification rate(%) Classification rate(%) Classification rate(%) Classification rate(%)

0
10
20
30
40
50
60
70
80
0
10
20
30
40
50
60
70
80
90
100
0
20
40
60
80
100
120
0
20
40
60
80
100
120

STP(1,2)
STP(1,2) STP(1,2) STP(1,2)
STP-UL(1) STP-UL(1) STP-UL(1) STP-UL(1)
STP-L(1)-L(2) STP-L(1)-L(2) STP-L(1)-L(2) STP-L(1)-L(2)
MNTCDP MNTCDP MNTCDP MNTCDP
ARCS-LBP ARCS-LBP ARCS-LBP ARCS-LBP
STP-L(2) STP-L(2) STP-L(2) STP-L(2)
MSJLBP MSJLBP MSJLBP MSJLBP
AHP AHP AHP AHP
ACS-LBP ACS-LBP ACS-LBP ACS-LBP
mdLBP mdLBP mdLBP mdLBP
Figure 9: (1):Brodatz ,(2):Jerry Wu and (3):Kylberg

MRELBP MRELBP MRELBP MRELBP

NI/RD/CI/LBP NI/RD/CI/LBP NI/RD/CI/LBP NI/RD/CI/LBP
DSLGS

Figure 11: (7):Curet ,(8):USPtex1-0 and (9):NewBarkTex

DSLGS DSLGS DSLGS

(9)
(6)
(3)

RSMAS
Kylberg

Figure 10: (4):XUH ,(5):KTH-TIPS and (6):KTH-TIPS2b

(12)

Figure 12: (10):Outex-TC13 ,(11):VisTex and (12):RSMAS

LQPAT LQPAT LQPAT LQPAT
KTH-TIPS2b

NewBarkTex

Descriptors
RD-CI-LBP

Descriptors
Descriptors
Descriprors

RD-CI-LBP RD-CI-LBP RD-CI-LBP

AECLBP-S AECLBP-S AECLBP-S AECLBP-S
LDTP LDTP LDTP LDTP
AECLBP-M AECLBP-M AECLBP-M AECLBP-M
LFDPhase LFDPhase LFDPhase LFDPhase
LCCP LCCP LCCP LCCP
LDGP LDGP LDGP LDGP
dLBPa dLBPa dLBPa dLBPa
ID-LBP ID-LBP ID-LBP ID-LBP
LDENP LDENP LDENP LDENP
VAR-LBP VAR-LBP VAR-LBP VAR-LBP
CRLBP-C CRLBP-C CRLBP-C CRLBP-C
 (a) (b) (c)
120 4000 12

3500
100 10

3000

The treatment time (in minutes)

80 8
Victories/comparisons

2500

Dimension
60 2000 6

1500
40 4

1000

20 2
500

0 0 0
BSIF

BSIF

MBP
STP-(1,2)

STP-U1U2

ILTP
LCxMSP

MNTCDP

LCvMSP

LTP
IBGC1

MSJLBP

AHP
CRLBPS

LBP
BGC1

MBP

STP-(1,2)
CLBPSMxC

ALBGC
ILBP
ELGS
AELTP

RCSLBP

nLBPd
LPQ
RILBP
maLBP

DCLBP
mdLBP

RLBP
MRELBP

ILTP
LCxMSP

MNTCDP

LTP
IBGC1

MSJLBP

AELTP
AHP
CRLBPS

LBP
BGC1

LPQ
RILBP

STP-(1,2)

ILTP

BSIF
LCxMSP

MNTCDP

LTP
IBGC1

MSJLBP

LBP
BGC1

DCLBP
MBP
STP-U1L1L2
STP-U1L1U2
STP-U1L1
STP-U1L1L2

STP-L1U2

STP-U1

STP-L1
RALBGC
ARCSLBP
STP-U2

STP-L2

LQP
RLBGC
SLBP
LDTP(2)

STP-L1
RALBGC
ARCSLBP
STP-U2

STP-L2

LQP
RLBGC
SLBP

ILBP
LDTP(2)
ALBGC

ELGS

RCSLBP

nLBPd

maLBP
mdLBP
DCLBP

RLBP
MRELBP

RCSLBP

nLBPd
LPQ
RILBP
maLBP
mdLBP

RLBP
MRELBP
STP-U1U2L2

STP-U1L2

STP-L1L2
LCCSMP
STP-U1L2

ACSLBP

STP-U1L1L2
STP-U1L1U2
STP-U1L1
STP-U1L1L2

STP-L1U2

STP-U1U2
STP-U1U2L2

STP-U1L2

STP-L1L2
LCCSMP

CLBPSMxC

LCvMSP
STP-U1L2

STP-U1

ACSLBP

STP-U1L1L2
STP-U1L1U2
STP-U1L1
STP-U1L1L2

STP-L1U2

STP-U1U2
STP-L1L2

CLBPSMxC

LCvMSP
STP-U1

STP-L1
RALBGC
ARCSLBP
STP-U2

STP-L2

LQP
RLBGC
SLBP
LDTP(2)
ALBGC
ILBP
ELGS
AELTP
AHP
CRLBPS
STP-U1U2L2

STP-U1L2

LCCSMP
STP-U1L2

ACSLBP
The first 50 texture descriptors tested The dimensions of the first 50 descriptors tested The first 50 texture descriptors tested

Figure 13: (a):The first 50 ranked tested texture descriptors ,(b):The dimensions of the first 50 ranked tested
texture descriptors (c):The processing time (in minutes) of the first 50 ranked tested texture descriptors.

320 [4] M. Worring, A. Smeulders, A. Gupta, S. Santini and R. Jain, ”Content-Based Image Retrieval at the End
321 of the Early Years” in IEEE Transactions on Pattern Analysis Machine Intelligence, vol. 22, no. 12, pp.
322 1349-1380, 2000. doi: 10.1109/34.895972
323 [5] F. Lehmann, “Turbo segmentation of textured images,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 33,
324 no. 1, pp. 16–29, Jan. 2011.
325 [6] M Pietikäinen, A Hadid, G Zhao, T Ahonen, Computer Vision Using Local Binary Patterns, vol. 40 of
326 Computational Imaging and Vision. (Springer, London, 2011)
327 [7] Haralick R. M., Shanmugan K., Dinsatein I., 1973: Textural Features for Image Classification. IEEE Trabs.
328 System, Man, and Cybernetics, vol. SMC-3, n°6, 610-621.
329 [8] X. Xie, ”A Review of Recent Advances in Surface Defect Detection using Texture analysis”, Electronic
330 Letters on Computer Vision and Image Analysis 7(3):1-22, 2008
331 l
332 [9] Ojala T, Pietikainen M, Maenpaa T (2002) Multiresolution gray-scale and rotation invariant texture classi-
333 fication with local binary patterns[J]. IEEE Trans Pattern Anal Mach Intell 24(7):971–987
334 [10] X. Tan, B. Triggs, Enhanced local texture feature sets for face recognition under dicult lighting conditions,
335 IEEE transactions on image processing 19 (6) (2010) 16351650.
336 [11] W. Yang, C. Sun, Face recognition using improved local texture patterns, in: Proceedings of the WCICA,
337 2011.
338 [12] Nanni, L., Lumini, A., Brahnam, S., 2010. Local binary patterns variants as texture descriptors for medical
339 image analysis. Artif. Intell. Med. 49 (2), 117–125, CrossRef.
340 [13] RSMAS- Shihavuddin A (2017) Coral reef dataset, v2. 10.17632/86y667257h.2file-5a2847d2-4c9f-41a9-
341 8d7ccdc74a0195c2
342

343 [14] BarkTex- A. Porebski, N. Vandenbroucke, L. Macaire, D. Hamad, “A new benchmark image test suite for
344 evaluating colour texture classification schemes”, Multimed. Tools. Appl. Springer, vol. 70(1), pp. 543-556,
345 2014.
346

347 [15] Fernández, A., Álvarez, M.X., Bianconi, F., 2013. Texture description through histograms of equivalent
348 patterns. J. Math. Imaging Vision 45, 76–102.
349 [16] [Link] merabet,[Link] Concave-and-Convex Micro-Structure Patterns for texture classification
350 Pattern Recognition on ScienceDirect Volume 76, April 2018, Pages 303-322
351 [17] LQCH- Zhao, Yang, Wang, Rong-Gang., Wang, Wen-Min, Gao, Wen, Local quantization code histogram
352 for texture classification. Neurocomputing[Link]
353 [18] CS-LMP- H. Zeng, X. Wang, Y. Gu, Center symmetric local multilevel pattern based descriptor and its
354 applicationin image matching, International Journal of Optics, dio: [Link]
355 v:2016.

21
356 [19] LESTP-Vipparthi, S.K., Murala, S., Nagar, S.K., Gonde, A.B., 2015. Local gabor maximum edge position
357 octal patterns for image retrieval. Neurocomputing 167, 336–345.
358 [20] LECTP Vipparthi, S.K., Nagar, S.K., 2015. Local extreme complete trio pattern for multimedia image
359 retrieval system. Int. J. Autom. Comput..
360 [21] AELTP- Kechen, Song, Yunhui, Yan, Yongjie, Zhao, et al., 2015. Adjacent evaluation of local binary
361 pattern for texture classification. J. Vis. Commun. Image Represent. 33, 323–339.
362 [22] CBP- Fu, X., Wei, W., 2008. Centralized binary patterns embedded with image euclidean distance for facial
363 expression recognition. In: Fourth International Conference on Natural Computation, vol. 4. pp. 115–119.
364 [23] CS-LBP Heikkilä, M., Pietikäinen, M., Schmid, C., 2006. Description of interest regions with centersym-
365 metric local binary patterns. In: Kalra, P.K., Peleg, S. (Eds.), 5th Indian Conference on Computer Vision,
366 Graphics and Image Processing, vol. 4338. LNCS,Madurai, India, pp. 58–69.
367 [24] (ICS-TS- CS-TS ) Zeng, H., 2011. A robust method for local image feature region description. Acta Autom.
368 Sin. 37 (6), 658–664.
369 [25] QBP- Zeng, H., Chen, J., Cui, X., Cai, C., Ma, K.K., 2016. Quad binary pattern and its application in
370 mean-shift tracking. Neurocomputing
371 [26] GTUC -Chang, C.I., Chen, Y.: Gradient texture unit coding for texture analysis. Opt. Eng. 43(8),
372 1891–1902 (2004)
373 [27] nLBP-nLBPa-Kaya, Y., Ertugrul, Ö.F., Tekin, R., 2015. Two novel local binary pattern descriptors for
374 texture analysis. Appl. Soft Comput. 34, 728–735
375 [28] El merabet, Y., Ruichek, Y., 2018. Local concave-and-convex micro-structure patterns for texture classifi-
376 cation. Pattern Recognit. 76, 303–322..
377

378 [29] CLBP-C Guo, Z., Zhang, L., Zhang, D.: A completed modeling of local binary pattern operator for texture
379 classification. IEEE Trans. Image Process. 19(6), 1657–1663 (2010)
380 [30] [Link],[Link] merabet,[Link],[Link] neighborhood topology cross decoded patterns for
381 image-based face recognition, in: Expert Systems with Applications on ScienceDirect Volume 114, 30 De-
382 cember 2018, Pages 119-142
383 [31] RALBGC- El-khadiri, I., Kas, M., El merabet, Y., Ruichek, Y., Touahni, R., 2018b. Repulsive-
384 andattractive local binary gradient contours: New and efficient feature descriptors for texture classification.
385 Inform. Sci. [Link]
386 [32] Y. El merabet,Y. Ruichek,A. El idrissi Attractive-and-repulsive center-symmetric local binary patterns for
387 texture classification. Engineering Applications of Artificial Intelligence 78 (2019) 158–172
388 [33] BSIF- Juho Kannala and Esa Rahtu . Binarized Statistical Image Features. 21st International Conference
389 on Pattern Recognition (ICPR 2012) November 11-15, 2012. Tsukuba, Japan
390 [34] MSJLBP - Xianbiao Qi (BUPT), Yu Qiao (SIAT), Chun-Guang Li (BUPT), Jun Guo (BUPT) . Multi-
391 scale Joint Encoding of Local Binary Patterns for Texture and Material [Link] Proceedings British
392 Machine Vision Conference 2013. Pages 40.1–40.11 [Link]
393 [35] SLBP- Timo Ahonen, Matti Pietikainen. SOFT HISTOGRAMS FOR LOCAL BINARY PATTERNS Ma-
394 chine Vision Group 2007, Infotech Oulu PO Box 4500, FI-90014 University of Oulu, Finland,
395 [36] LDTP(2)- El-khadiri, I., Chahi, A., El-merabet, Y., Ruichek, Y., Touahni, R., 2018a. Local directional
396 ternary pattern: A new texture descriptor for texture classification. Comput. Vis. Image Underst. 169,
397 14–27. [Link]
398 [37] ILBP- Wei Yu · Lin Gan · Sha Yang · Yonggang Ding Pan Jiang · Jun Wang · Shijun Li ,An improved
399 LBP algorithm for texture and face classification. SIViP DOI 10.1007/s11760-014-0652-5
400 [38] ELGS- Bashier, H.K., Hoe, L.S., Hui, L.T., Azli, M.F., Han, P.Y., Kwee, W.K., Sayeed, M.S., 2016. Texture
401 classification via extended local graph structure. Optik 127, 638–643.
402 [39] AHP- Zhu, Z., You, X., Chen, C.P., Tao, D., Ou, W., Jiang, X., Zou, J., 2015. An adaptive hybrid pattern
403 for noise-robust texture analysis. Pattern Recognit. 48 (8), 2592–2608.

22
404 [40] CRLBP- Yang Zhao a,Wei Jia,Rong-Xiang Hu,Hai [Link] robust local binary pattern for texture
405 [Link] 106 (2013) 68–76
406 [41] LPQ- Ojansivu, V., Heikkila, J., 2008. Blur insensitive texture classification using local phase quantization.
407 In: Elmoataz, A., Lezoray, O., Nouboud, F., Mammass, D. (Eds.), Image and Signal Processing, vol. 5099.
408 LNCS, pp. 236–243.
409 [42] RILBP- Lin, F.-Y., Zheng, C.-H., Wang, X.-F., Man, Q.-K., 2008. Multiple classification of plant leaves
410 based on gabor transform and lbp operator. International Conference on Intelligent Computing. Springer,
411 pp. 432–439.
412 [43] maLBP-mdLBP- Dubey, S.R., Singh, S.K., Singh, R.K., 2016b. Multichannel decoded local binary patterns
413 for content based image retrieval. IEEE Trans. Image Process. 25 (9), 4018–4032.
414 [44] DCLBP- Ylioinas, J., Hadid, A., Guo, Y., Pietikäinen, M., 2012. Efficient image appearance description
415 using dense sampling based local binary patterns. In: Asian conference on computer vision. Springer, Berlin,
416 Heidelberg, pp. 375–388.

417 [45] MBP- A. Ha

418 ane, G. Seetharaman, B. Zavidovique, Median binary pattern for textures classi
419 cation, in: 475 International Conference Image Analysis and Recognition, Springer, 2007, pp. 387398.
420 [46] RLBP -J. Chen, V. Kellokumpu, G. Zhao, M. Pietikainen, Rlbp: Robust local binary pattern., in: British
421 Machine Vision Conference (BMVC), 2013.

422 [47] MRELBP- Liu, L., Lao, S., Fieguth, P.W., Guo, Y., Wang, X., Pietikäinen, M., 2016. Median robust
423 extended local binary pattern for texture classification. IEEE Trans. Image Process. 25 (3), 1368–1381.
424 [48] FDLBP - 2017. Face retrieval using frequency decoded local descriptor. arXiv:1709.06508. Faraji, M.R., Qi,
425 X., 2016. Face recognition under varying illuminations using logarithmic fractal dimension-based complete
426 eight local directional patterns. Neurocomputing 199, 16–30.
427 [49] LBPHF- Ahonen, T., Matas, J., He, C., Pietikäinen, M., 2009. Rotation invariant image description with
428 local binary pattern histogram Fourier features. In: Scandinavian Conference on Image Analysis. Springer,
429 Berlin, Heidelberg, pp. 61–70.
430 [50] LNDP- Verma, M., Raman, B., 2017. Local neighborhood difference pattern: A new feature descriptor for
431 natural and texture image retrieval. In: Multimedia Tools and Applications. pp. 1–24.
432 [51] NI-RD-LBP- Liu, L., Zhao, L., Long, Y., Kuang, G., Fieguth, P., 2012. Extended local binary patterns for
433 texture classification. Image Vis. Comput. 30 (2), 86–99.
434 [52] LFDMag- Rouzbeh Maani a,n, SanjayKalra b,c, Yee-HongYang Noise robust rotaion features for classifi-
435 cation .Pttern Recognition 46(2013)2103–2116
436 [53] LGS- E. E. A. Abusham and H. [Link], “Face recognition using local graph structure (LGS),” Lecture
437 Notes in Computer Science, vol. 6762, pp.169–175, 2012.
438 [54] 3DLBP-Huang, Y., Wang, Y., Tan, T.: Combining statistics of geometrical and correlative features for 3D
439 face recognition. In: Proceedings of the 17th BritishMachine Vision Conference,2006. pp. 879–888 (2006

440 [55] LMEBP- Subrahmanyam, M., Maheshwari, R.P., Balasubramanian, R., 2012. Local maximum edge binary
441 patterns: A new descriptor for image retrieval and object tracking. Signal Process. 92 (6), 1467–1479.
442 [56] SLGS Song Dong, Jucheng Yang, Chao Wang, Yarui Chen, Di Sun. A New Finger Vein Recogni-
443 tion Method Based on the Difference Symmetric Local Graph Structure (DSLGS).International Jour-
444 nal of Signal Processing, Image Processing and Pattern Recognition Vol.8, No.10 (2015), pp.71-80
445 [Link]
446 [57] DSLGS- Dong, S., Yang, J., Wang, C., Chen, Y., Sun, D., 2015. A new finger vein recognition method
447 based on the Difference Symmetric Local Graph Structure, DSLGS. Int. J. Signal Process., Image Process.
448 Pattern Recognit. 8 (10), 71–80.

449 [58] LOOP- Tapabrata Chakraborti , Brendan McCane , Steven Mills, and Umapada Pal. LOOP Descriptor:
450 Local Optimal-Oriented [Link] SIGNAL PROCESSING LETTERS, VOL. 25, NO. 5, MAY 2018

23
451 [59] LQPAT- Chakraborty, S., Singh, S.K., Chakraborty, P., 2017c. Local quadruple pattern: A novel descriptor
452 for facial image recognition and retrieval. Comput. Electr. Eng..
453 [60] WLD - Jie Chen, , Shiguang , Chu He, Guoying Zhao, Matti Pietika¨ inen, , Xilin Chen, , IEEE, and Wen
454 Gao. WLD: A Robust Local Image [Link] TRANSACTIONS ON PATTERN ANALYSIS AND
455 MACHINE INTELLIGENCE, VOL. 32, NO. 9, SEPTEMBER 2010 1705
456 [61] SMEPOP- Vipparthi, S.K., Murala, S., Nagar, S.K., Gonde, A.B., 2015. Local gabor maximum edge
457 position octal patterns for image retrieval. Neurocomputing 167, 336–345.
458 [62] DRLBP - Mehta, R., Egiazarian, K., 2016. Dominant Rotated Local Binary Patterns, DRLBP for texture
459 classification. Pattern Recognit. Lett. 71, 16–22.
460 [63] MD-LBP- Jin Liu1 ,Yue Chen1 ,Shengnan Sun1 A novel local texture feature extraction
461 method called multi-direction local binary [Link] Tools and Applications-2019.
462 [Link]
463 [64] AELBP- Kechen, Song, Yunhui, Yan, Yongjie, Zhao, et al., 2015. Adjacent evaluation of local binary
464 pattern for texture classification. J. Vis. Commun. Image Represent. 33, 323–339.
465 [65] CSQP- Chakraborty, S., Singh, S.K., Chakraborty, P., 2017a. Centre symmetric quadruple pat-
466 tern: A novel descriptor for facial image recognition and retrieval. Pattern Recognit. Lett.
467 [Link]
468 [66] STS- B. Xu, P. Gong, E. Seto, S. R, Comparison of gray-level reduction and dierent texture spectrum
469 encoding methods forland-use classi
470 cation using a panchromatic ikonos image, Photogrammetric Engineering Remote Sensing 69 (2003) 529536.
471 [67] LDTP Rivera, A.R., Castillo, J.R., Chae, O., 2015. Local directional texture pattern image descriptor.
472 Pattern Recognit. Lett. 51, 94–100.
473 [68] CCR- E. Kurmyshev, M. Cervantes, A quasi-statistical approach to digital binary image representation,
474 Revista Mexicana de Fsica 42 (1) (1996) 104116.
475 [69] CELDP- Faraji, M.R., Qi, X., 2016. Face recognition under varying illuminations using logarithmic fractal
476 dimension-based complete eight local directional patterns. Neurocomputing 199, 16–30.
477 [70] DCP- Ding, C., Choi, J., Tao, D., Davis, L.S., 2016. Multi-directional multi-level dual-cross patterns for
478 robust face recognition. IEEE Trans. Pattern Anal. Mach. Intell. 38 (3), 518–531.
479 [71] LDP- T. Jabid, M. H. Kabir, O. Chae, Local directional pattern (ldp) for face recognition, in: Proceedings
480 of the IEEE International Conference on Consumer Electronics, 2010, pp. 329330.
481 [72] CBP- X. Fu, [Link], Centralized binary patterns embedded with image euclidean distance for facial ex-
482 pression recognition, in: Natural Computation, 2008. ICNC’08. Fourth International Conference on, Vol. 4,
483 IEEE, 2008, pp. 115119.
484 [73] LFDPhase- Rouzbeh Maani, Sanjay Kalra, and Yee-Hong [Link] Invariant Local Frequency De-
485 scriptors for Texture Classification IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 22, NO. 6,
486 JUNE 2013
487 [74] GTUC- Chang, C.I, Y. Chen, Gradient texture unit coding for texture analysis, Optik 43 (8) (2004)
488 18911902.
489 [75] LDN- Rivera, A.R., Castillo, J.R., Chae, O.O., 2013. Local directional number pattern for face analysis:
490 Face and expression recognition. IEEE Trans. Image Process. 22 (5), 1740– 1752.
491 [76] GLBP- He, Y., Sang, N.: Robust illumination invariant texture classification using gradient local binary
492 patterns. In: Proceedings of 2011 InternationalWorkshop on Multi-Platform/Multi-Sensor Remote Sensing
493 and Mapping, Xiamen, China, pp. 1–6 (2011)
494 .
495 [77] LDRP- 2017. Local directional relation pattern for unconstrained and robust face retrieval.
496 arXiv:1709.09518.
497 [78] ELDP- Faraji, M.R., Qi, X., 2015. Face recognition under illumination variations based on eight local
498 directional patterns. IET Biom. 4 (1), 10–17

24
499 [79] LCCP- X. Chen, Z. Zhou, J. Zhang, Z. Liu, Q. Huang, Local convex-and-concave pattern: An eective
500 texture
501 descriptor, Information Sciences 363 (2016) 120139.
502 [80] SME-MME-POP- Vipparthi, S.K., Murala, S., Nagar, S.K., Gonde, A.B., 2015. Local gabor maximum
503 edge position octal patterns for image retrieval. Neurocomputing 167, 336–345.
504 [81] D-LBP-Wu, X., Sun, J., 2009. An effective texture spectrum descriptor. Information Assurance and Secu-
505 rity, 2009. IAS’09. Fifth International Conference on. 2. IEEE, pp. 361–364.
506 [82] XCSLBP- C. Silva, T. Bouwmans, C. Frelicot, An extended center-symmetric local binary pattern for
507 background modeling and subtraction in videos, in: International Joint Conference on Computer Vision,
508 Imaging and Computer Graphics Theory and Applications, VISAPP 2015, 2015.
509 [83] DC- Farida Ouslimani , Achour Ouslimani, Zohra Ameur. Rotation-invariant features based on directional
510 coding for texture [Link] Computing and Applications [Link]
511 3462-9(0123456789().,-volV)(0123456789().,-volV)

512 [84] LDGP- Chakraborty, S., Singh, S.K., Chakraborty, P., 2015. Local directional gradient pattern: a local
513 descriptor for face recognition. Multimedia Tools Appl. 76 (1), 1201–1216.
514 [85] MST- Xu, B., Gong, P., Seto, E., Spear, R., 2003b. Comparison of gray-level reduction and different texture
515 spectrum encoding methods for land-use classification using a panchromatic ikonos image. Photogramm.
516 Eng. Remote Sens. 69 (5), 529–536.

517 [86] LWP- Dubey, S.R., Singh, S.K., Singh, R.K., 2015a. Local wavelet pattern: A new feature descriptor for
518 image retrieval in medical CT databases. IEEE Trans. Image Process. 24 (12), 5892–5903
519 [87] LDEP- Dubey, S.R., Singh, S.K., Singh, R.K., 2015b. Local diagonal extrema pattern: A new and efficient
520 feature descriptor for CT image retrieval. IEEE Signal Process. Lett. 22 (9), 1215–1219

521 [88] GLTCS+- Hepplewhite, L., Stonhamm, T.J.: Texture classification using N-tuple pattern recognition. In:
522 Proceedings of the 13th International Conference on Pattern Recognition (ICPR’96), vol. 4, pp. 159–163
523 (1996). doi:10.1109/ICPR.1996.547253
524 .
525 [89] LDNP- Adin Ramirez Rivera, Student Member, IEEE, Jorge Rojas Castillo, Student Member, IEEE, and
526 Oksam Chae*, Member, IEEE . Local Directional Number Pattern for Face Analysis: Face and Expression
527 RecognitionIEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 22, NO. 5, YEAR 2013
528 [90] ELDP- Zhong, F., Zhang, J., 2013b. Face recognition with enhanced local directional patterns. Neurocom-
529 puting 119, 375–384.

530 [91] OTF- Xu, Y., Huang, S., Ji, H., Fermüller, C., 2012a. Scale-space texture description on SIFT-like textons.
531 Comput. Vis. Image Underst. 116 (9), 999–1013.
532 [92] RT- Zabih, R.,Woodfill, J.: Non-parametric local transforms for computing visual correspondence. In:
533 Proceedings of the 3rd European Conference on Computer Vision (ECCV 1994), Stockholm, Sweden, pp.
534 151–158. Springer, Berlin (1994)

535 [93] LBDP Dubey, S.R., Singh, S.K., Singh, R.K., 2016a. Local bit-plane decoded pattern: A novel feature
536 descriptor for biomedical image retrieval. IEEE J. Biomed. Health Inf. 20 (4), 1139–1147
537 [94] ID-LBP- Wu, X., Sun, J., 2009. An effective texture spectrum descriptor. Information Assurance and
538 Security, 2009. IAS’09. Fifth International Conference on. 2. IEEE, pp. 361–364.

539 [95] BTCS+- Patel, D., Stonham, T., 1991. A single layer neural network for texture discrimination. Circuits
540 and Systems, 1991., IEEE International Sympoisum on. IEEE, pp. 2656–2660.
541 [96] LBC- Zhao, Y., Huang, D.-S., Jia, W., 2012. Completed local binary count for rotation invariant texture
542 classification. IEEE Trans. Image Process. 21 (10), 4492–4497.
543 [97] LDENP- Pillai, A., Soundrapandiyan, R., Satapathy, S., Satapathy, S.C., Jung, K.H., Krishnan, R., 2017.
544 Local diagonal extrema number pattern: A new feature descriptor for face recognition. Future Gener.
545 Comput. Syst. [Link] 09.055.

25
546 [98] LOSIB- Garca-Olalla, O., Alegre, E., Fernández-Robles, L., González-Castro, V., 2014. Local Oriented
547 Statistics Information Booster LOSIB for texture classification. In: International Conference on Pattern
548 Recognition. pp. 1114–1119..
549 [99] FLBP- Mohammadi, M.R., Fatemizadeh, E., 2013. Fuzzy local binary patterns: a comparison between
550 min-max and dot-sum operators in the application of facial expression recognition. Machine Vision and
551 Image Processing (MVIP), 2013 8th Iranian Conference on. IEEE, pp. 315–319.
552 [100] VAR-LBP- Guo, Z., Zhang, L., Zhang, D., 2010b. Rotation invariant texture classification using lbp
553 variance (lbpv) with global matching. Pattern Recognit 43 (3), 706–719.
554 [101] LBPD- Hong, X., Zhao, G., Pietikäinen, M., Chen, X., 2014. Combining LBP difference and feature
555 correlation for texture description. IEEE Trans. Image Process. 23 (6), 2557–2568
556 [102] AECLBP-C -Kechen, Song, Yunhui, Yan, Yongjie, Zhao, et al., 2015. Adjacent evaluation of local binary
557 pattern for texture classification. J. Vis. Commun. Image Represent. 33, 323–339.
558 [103] CRLBP-C- Yang Zhao , WeiJia , Rong-XiangHu, HaiMin .Complete robust local binary pattern fortexture
559 classification. Neurocomputing 106(2013)68–76

26