Gradient-Based Feature Extraction From Raw Bayer Pattern Images

In this paper, the impact of demosaicing on gradient extraction is studied and a gradient-based feature pipeline based raw Bayer pattern images proposed. It shown both theoretically experimentally that are applicable to central difference algorithms with negligible performance degradation, as long arrangement color filter array (CFA) patterns matches operators. The constancy assumption, which widely used in various algorithms, applied proposed image-based pipeline. Experimental results show gradients extracted from robust enough be histogram oriented (HOG)-based pedestrian detection shift-invariant transform (SIFT)-based matching algorithms. By skipping most steps image signal processing (ISP) pipeline, computational complexity power consumption computer vision system can reduced significantly.