Scribd
Upload
84 views4 pages

Procesamiento Digital de Imágenes

The document outlines 25 suggested exercises for an introductory digital image processing course, including tasks to manipulate image resolution, dynamic range, color channels, contrast, noise, filters, transforms, restoration, and segmentation. Students are evaluated based on exams on lecture presentations, applying ImageJ routines to images, and an image processing project in experimental physics. The exercises cover a range of fundamental image processing techniques but students need not complete all of them.

Uploaded by

api-26891612
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
84 views4 pages

Procesamiento Digital de Imágenes

The document outlines 25 suggested exercises for an introductory digital image processing course, including tasks to manipulate image resolution, dynamic range, color channels, contrast, noise, filters, transforms, restoration, and segmentation. Students are evaluated based on exams on lecture presentations, applying ImageJ routines to images, and an image processing project in experimental physics. The exercises cover a range of fundamental image processing techniques but students need not complete all of them.

Uploaded by

api-26891612
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
You are on page 1/ 4

Digital Image Processing I. An Intro.

I) Suggested exercises. Use the programming language of your choice (there


aren’t many indeed)
MATLAB, IDL, DELPHI). Those that cannot program using a high level language
will assist to an interactive IDL class (2hrs/week) while the power point
presentations are discussed in class. Remember that you will use known images
to be able to make comments on a particular processing technique to be able to
correctly assess your results.

1.) Spatial resolution: Display at different resolutions a black and white (B/W)
image of your choice. Use Barbara , Elaine, Lenna, Valerie or whatever suits
your mood!.
Display the image in 512x512 (or starting with the max resolution), 256x256,
128x128, 64x64, 32x32. Display each image at the same physical size on the
CRT screen.

2) . Dynamic Range: Choose a B/W image and display it, varying the dynamic
range from 2 bits (binary, two gray levels), 3 bits up to 8 bits (256 gray levels).
Remember that B/W images are in general displayed only with 256 gray levels.

3) . Choose an image F ( m, n) and display it in 256 gray levels. Map it from


[0,255] into [0,1]. Modify the image logarithmically

ln[1 + F (m, n)]


G (m, n) =
ln(2)
Quantize the above image G (m, n) to a given number of gray levels as done in
exercise (2) (use e.g. 8) to produce Gˆ (m, n) (for example 8 levels). In the next
{ G ( m,n ).ln(2)}
ˆ
step generate the image Fˆ (m, n) = e −1
Compare the result to F (m, n) quantized to the same number of gray levels as
performed in (2) and display them side by side. Make the comments concerning
your observations.

4) . Choose an RGB image from the “collection” folder (Lenna, Girl_pratt,


Baboon, peppers, Tiffany, Lara, bird etc…). Split the image into 3 channels (R,G
and B). Generate “false” color images using all possible combinations.
5). Choose a specified image from the collection folder which has poor contrast.
Check out Min and Max of that image. Stretch the dynamic range of the image to
the range [0,255]. Choose a true color image with poor contrast and stretch
individually ([min,max], flexible) the 3 color planes and recombine them to a true
color image.
Write a piece of code that rotates an image by an arbitrary angle ϕ .
Write a piece of code that performs zooming of some section of an image.

6.) Perform histogram equalization on a B/W image with poor contrast and on
another one with normal contrast (if you choose to use a color image then
perform it on each plane and see what you get) Otherwise chose any B/W image
from the collection. Make some comments. Perform now histogram equalization
on three planes of some true color image and recombine to a true color image.
Compare side by side the original color image and the color histogram equalized
image. Use the Tahoe tree in the snow though it should not be done!) or the
girl2_contrast image for example, stretch the histogram as in 5) and compare to
a plane by plane histogram equalized image.

7) . Perform Image enhancement. Apply various masks to some image of your


choice. Use the convolution routine of the corresponding toolbox of your
language. Use different noise cleaning masks. Use “North”, NW, W, SW, S, SE,
E, NE masks and apply a threshold and display the threshold image.
Perform edge crispening or sharpening using unsharp masking. Use high-pass
masks to perform the same and compare the results.
Apply the Marr-Hildreth sombrero filter at different scales to make an edge
detection program.
Use some face to detect the edges and visually the result according to your
knowledge.
Improve the “Sharon.png” picture.
Process optical illusion images. Make some comments.
Count the balls in test3.png and jelly_beans.png

8). Enhance the B/W bridge image using the statistical differencing technique.
Apply the same technique to 010baker, afterthoughts(cheetah), asp, face_gal,
Natalie_512,Julie_in_dark,unknown pictures.

9). Use the statistical differencing from Wallis to enhance the same images from
above. Use the parameters given in class.

10). Apply the median filter to an image to eliminate the discrete impulse noise.
11). Perform multispectral image enhancement: use contrast stretching,
histogram equalization and edge crispening as used above applying it to the
spectral channels before recombining them.

12) . When an image plane obtained from a multispectral imaging system can be
modeled as the product of an illumination function and a reflectivity function,
Fn (m, n) = E (m, n) Rn (m, n) then, building ratios of the different spectral planes
Fm (m, n)
enhances the gray scale quantization noise. This can be reduced by
Fn (m, n)
taking the logarithm of each spectral plane written as the above product and then
performing the ratioing operations.

13) . Choose some B/W image and reduce it to 64x64 image (use radix 2).
Implement the discrete Fourier transform ad hoc by computing directly the
double sum involved. Use the FFT algorithm and compare the time savings.

14 ) . Use ideal low-pass, band-pass and high-pass filters and apply them to
some B/W image. Observe the effects and make comments.

15). Perform the same as above but this time using a Gaussian type low-pass
and band-pass filter.

16). Perform edge detection using simple masks and make some comments.
See masks in class.

17). Image restoration: use an image and blur it slightly with some predefined
Gaussian function for example. Add to it very moderate noise. You can generate
it with a random generator and add it to the blurred image. Now perform
deconvolution or image restoration using Fourier space techniques compare the
result to the original unblurred image and make your comments.

18) . Perform your own image compression using the discrete cosine transform
which yields the JPEG format. Use 8x8 blocks and retain a variable number of
cosine coefficients to reconstruct the compressed image.

19) . The same as above can be done using wavelets applied to local
subimages. This should yield much better results.

20) . Perform a simple segmentation task on a specially chosen image of course.


Use the k-means algorithm described in class. Implement it and use a variable
number of clusters. Compare the results and make comments. Can we apply it
meaningfully to any image, for example the baboon ?
21). Inverse Filtering: take a high resolution image, blur it with a known PSF and
then perform inverse filtering to recover the original image. Add progressively
random noise and perform again the task of image restoration using the same
blur function. Observe how the process complete fails once some noise level has
been reached.

22.) Image restoration task: use a Wiener filter to restore the image of the above
exercise #21. Estimate the power spectrum of the original image from the power
spectrum of the degraded blurred image.

23.) Repeat the task of exercise #22 using homomorphic filtering technique to
perform digital image restoration. Estimate again the power spectrum as above.

24.) Use a motion blurred image to restore it using the SVD restoration
technique. Include successively smaller and smaller eigenvalues of the SVD
decomposed PSF.

25.) write a routine to rotate an image by an arbitrary angle theta. Apply first a
circular mask to any face or object.

Second grade
II) You will have one exam with questions about the presentations.

Third grade
III) you will have to apply most of the ImageJ routines to some images and write
it up.

Fourth grade
IV) Project: Use your image processing skills and mathematical skills to apply
them to specific projects of experimental Physics (see non-linear Physics Lab)

You will not have to solve all suggested exercises; you won’t have the time for it.
You final grade depends largely on your personal efforts on getting through the 4
principal topics I)…IV)

You might also like