Feature Detection Syllabus Edge detection, corner detection, line and curve detection, active contours, SIFT and HOG descriptors, shape context descriptors, Morphological operations Contents 3.1 Feature Detection 3.2 SIFT Descriptors and HOG Descriptors 3.3. Morphological Operations @-7) Scanned with CamScannerSomputor Vision eed Fontan ty, $8 tir 13.1] Feature Detection 4 feature is a piece of information about the content of an image in compute, and image proces i a‘ Be J Hain attributes. Specific structures in the image, such as points, edges, oF object, mi be used as fo, ‘atures. A generic neighborhood procedure or feature detection pertorrn, could potentially result in features. Other elements include motion jp ; S Well as shaj Sections, A fea ture, Ng, usually concerning whether a certain portion of the ip, i ; dt, the image sequences - ae lage pes defined by curves or the boundaries between distingy ima, « broader sense, is any piece of data that is useful to complet computing job associated w the features used ith a certain application. The feature idea is fairly broag | } { @ computer vision system might vary greatly depending «,, | situation at hand. When features are defined in terms of local neighborhood operations appl image, a procedure known as feature extraction, it is possible to distinguish between | feature detection approaches that produce local decisions about whether or not a given type of feature exists ata given image point, and those that produce non-binary data asa result. Feature detection is a low-level image processing operation. That is, it is usually | performed as the first operation on an image, and examines every pixel to see if there is; feature present at that pixel. If this is part of a larger algorithm, then the algorithm will typically only examine the image in the region of the features. As a built-in Pre-requisite to feature detection, the input image is usually smoothed by a Gaussian kernel in a scale- Space representation and one or several feature images are computed, often expressed in terms of local image derivatives operations. Methods for computing, abstractions of image information and making local choices whether or not there is an image feature of a certain type at each picture point are included in feature detection. The generated features will frequently take the shape of single dots, continuous curves, or connected regions, and will represent subsets of the image domain. Edge Detection dge detection refers to a set of mathematical techniques for recognizing points in 2 al image where the image brightness abruptly changes or, more formally, where there are discontinuities. The sharp fluctuations in image brightness are usually groupe! into a collection of curved line segments called edg The goal of detecting sharp changes in image brightness is to record significant events and changes in the worlds TECHNICAL PUBLICATIONS® - an up-thrust for knowledge Scanned with CamScanner3- computor visi’ 3 Feature Detection es. It may be demonstrated that, assuming very broad assumptions for an image rib A RNa 7 del, picture brightness discontinuities are likely to correlate to : ation Mo e Discontinuities in depth, «. Discontinuities in surface orientation, ‘Changes in material properties and «Variations in scene illumination. For picture segmentation, there are numerous edge detection algorithms in the jorature, This section examines the most widely used discontinuity-based edge detection approaches. Roberts edge detection, Sobel edge detection, Prewitt edge detection, Kirsh edge detection, Robinson edge detection, Marr-Hildreth edge detection, LoG edge etection, and Canny edge detection are some of the approaches available. Roberts edge detection Lawrence roberts is the inventor of the roberts edge detection (1965). It computes a simple 2-D spatial gradient measurement on an image in a short amount of time. This approach highlights high spatial frequency regions, which frequently correspond to edges. The most common application of this technology is to give the operator a grayscale image that matches the output. Every pixel value in the output represents the estimated entire magnitude of the input image's spatial gradient at that place. -1 0 0 =4 Sobel edge detection Sobel proposed the Sobel edge detection method in 1970 (Rafael C.Gonzalez, 2004). ‘The Sobel method of image segmentation edge detection uses the Sobel approximation to the derivative to find edges. At the points where the gradient is the highest, it comes before the edges. The Sobel approach emphasizes regions of high spatial frequency that correlate to edges by performing a 2-D spatial gradient quantity on an image. It's typically used to calculate the estimated absolute gradient magnitude at each location ina srayscale image. It's typically used to calculate the estimated absolute gradient magnitude at each location in a grayscale image. At the very least, the operator is made TECHNICAL PUBLICATIONS® - an up-thrust for knowtodge Scanned with CamScanner—— he table below. One ketne} 4 Shs © this, 9-4 ls, as shown in th sy is extremet; Me pets eross operator is extremely sig Computer Viewer —_ x J complication ke er. The up of a pair of 3 90 degree rotated front the oth Marr-Hildreth edge detection The Marr-Hildreth (1980) methodology is continuous curves methodology 5. detecting edges in digital images where there are well-built and quick fluctuations j, 's simple and works by convolving the image with the LoG Function, pid approximation. The edges are then detected by looking for er). ructure when turned upside-down, image brightness. or using DoGs as a raj d output. Because of its image st crossings in the filtere hat wavelet. The Marr-Hildreth ed». the LoG technique is also known as the Mexican detector has the following algorithm : + Apply a Gaussian smoothing filter to the image. Smooth the image with a two-dimensional Laplacian (often the first two steps are combined into a single operation) «Look for sign changes as you loop over the results. Mark as an edge if there isa sign change and the slope across the sign change is greater than a certain threshold. Although this is not how the edge detector was originally constructed, it is feasible to improve results by running the Laplacian result through a hysteresis similar ‘9 Canny's edge detection. LOG edge detection The Laplacian of Gaussian (LoG) was proposed by Marr(1982). The LoG of an imazt {(x, y) is a second order derivative defined as, a a Vie 4 dx? ay? TECHNICAL PUBLICATIONS® - an up-thrust for knowledge Scanned with CamScannerstil 3-5 tor Vision conto ~gnostw effects it smoothes the image and it computes the Laplacian, whch yields a \e-cge image, Locating edges then consists of Finding the zno crossings econ Jer quble edges. The digital implementation of the Laplacian function is commonly ; ed through the mask below, : Feature Detection the pecs’ The Laplacian is generally used to found whether a pixel is on the dark or light side of anedge- Canny edge detection The Canny edge detection technique is a standard edge detection technique in industry. It was developed by John Canny for his Master's thesis at MIT in 1983, and it still outperforms many of the more recent algorithms. Canny is a highly essential method for finding edges by isolating noise from the image before finding picture edges. Canny is a highly essential method for finding edges by isolating noise from the image before finding picture edges. The Canny approach is a better method since it Preserves the features of the image's edges after applying the tendency to detect the edges and a substantial threshold setting. The following are the algorithmic steps : * Convolve image f(r, c) with a Gaussian function to get smooth image (c, ©). Ac, ©) = £(r,c)"G(r,¢,6) + Compute edge strength using the difference gradient operator first, then compute edge magnitude and direction as before. * Suppress the gradient magnitude with non-maximal or critical suppression. * Apply a threshold to the image with non-maximal suppression. The Canny operation, unlike Roberts and Sobel, is not very sensitive to noise. It would be superior if the Canny detector worked properly. TECHNICAL PUBLICATIONS® - an up-thrust for knowodge Scanned with CamScanner