作者: Philipp Fischer , Thomas Brox , None
DOI: 10.1007/978-3-319-11752-2_19
关键词:
摘要: Descriptors based on orientation histograms are widely used in computer vision. The spatial pooling involved these representations provides important invariance properties, yet it is also responsible for the loss of details. In this paper, we suggest a way to preserve details described by local curvature. We propose descriptor that comprises direction and magnitude curvature naturally expands classical like SIFT HOG. demonstrate general benefit expansion exemplarily image classification, object detection, matching.