Hog feature extraction of------------image of graduation Project



HOG, a gradient-oriented histogram (histograms of oriented Gradient), was first proposed by Navneet Dalal and Bill Triggs in 2005, initially for pedestrian detection. They extracted HOG features and used linear SVM classifier to detect pedestrians, and achieved good results. The HOG characteristic is put forward on the assumption that the edge information and the gradient value distribution of the object's local range pixels can fully describe the shape and appearance of the object. The HOG feature is developed on the basis of SIFT image processing method, edge direction histogram and other methods, but there are some differences with these methods, the main manifestation is that the calculation of HOG features is confined to a uniformly distributed unit, and it also combines these individual units into overlapping blocks. To improve the performance of the algorithm.

before calculating the HOG features, the basic units and parameters used in the calculation process are introduced first. The sample resolution selected in this article is 64x64, which will also be the size of the inspection window for subsequent vehicle verification. The detection window is divided into blocks, each block is 16x16, and a block is divided into 4 8x8 cell,block with cell side length as the sliding step. Each cell's gradient histogram is divided into 9 bins, that is, each bin is 20 degrees, the gradient direction of the pixels in this range is delineated in the same class, with recoupling equal to the 0~180度 and 180 degrees ~360 degree of consolidation processing. Figure 4-6 detects the relationship between the window, block, and cell.

Comments:

we divide the image into several cells Cell", for example, each Cellto be6*6a pixel. Suppose we adopt9abinthe histogram to count this6*6the gradient information for each pixel. That isCellthe gradient direction thedegree into9direction block,: For example: If the gradient direction of this pixel is20-40degree, histogram section2abinThe count is added one, so that theCellThe weighted projection of each pixel in the histogram (mapped to a fixed angle range) in the gradient direction, you can get thisCellhistogram of the gradient direction, which is theCellcorresponding to the9Dimension eigenvectors (because there are9abin).

Hog feature extraction of------------image of graduation Project