HOG

HOG Descriptor

Histogram of oriented gradients (HOG) are feature descriptors for the object detection purpose. Descriptor is a set of feature vectors. Each feature vector is computed from a block placed across the source image. Each element of a vector is a histogram of gradient orientations.

 

The algorithm of finding the HOG descriptor consists of the following steps:

1. 1，Compute gradients for each pixel of an image.
2. 2，Perform binning of gradients orientation (from 0 to 180 degrees, opposite directions count as the same).

3，Collect the histogram within a cell of pixels.

1. 4，Weight the histogram by blocks and cells for local normalization of the contrasts.
2. 5，Normalize the histogram.
3. 6，Train a linear support vector machines (SVM) to detect an object. The output from the trained linear SVM is a set of coefficients for each element in a feature vector.

HOG

原文：http://www.cnblogs.com/sprint1989/p/3949600.html