3.2 HOG FEATURE
The gradient of the image, i.e., the differentials in the x and y directions, is very useful. The edges and corners have very large gradients. Compared with other areas on the plane, these areas with sharp intensity changes contain rich information on object shape. Thus, this paper designs and describes an HOG algorithm. For clarity, the steps of the algorithm were explained with an ordinary image with movement trends.
Preprocessing
The aspect ratio must be fixed before analyzing the multi-scale image blocks. Here, the aspect ratio of image blocks is maintained at 1:2. An original image of 720×475 resolution (figure 10) was selected as an example to illustrate HOG calculation. The small blocks cropped from the image were of the size 64×128.
Share with your friends: |