Histogram of Oriented Gradients Feature Extraction Without Normalization
Ling Zhang, Wei Zhou, Jingwei Li, Juan Li, Xin Lou
Abstract
In this paper, the effects of normalization in the histogram of oriented gradients (HOG) are studied and a HOG feature extraction pipeline without normalization is proposed. In the proposed pipeline, the functionality of normalization is merged into the gradient generation step by replacing the original linear difference based gradients with logarithmic gradients. Due to the discrete property of the pixel values, the logarithmic operation can be easily implemented using a lookup table (LUT) with a depth of 2 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">N</sup> , where N is the bit-width of the pixels. Theoretical analysis and experimental results show that the proposed normalization-free HOG feature based logarithmic gradient is close to the original version and can be used in the pedestrian detection algorithms without performance degradation. It is shown in the experiments that by skipping the time-consuming normalization step, the processing speed of HOG feature extraction can be significantly improved.