Akademik Veri Yönetim Sistemi
Eskişehir Technical University Journal of Science and and Technology A- Applied Sciences and Engineering, cilt.19, sa.3, ss.784-795, 2018 (Hakemli Dergi)
Texture recognition is one of the active fields in pattern recognition. Researchers have been searching for the best representationof a texture image for decades. The majority of methods use appearance-based properties of texture images to generate a featuredescriptor. In this paper, we propose novel feature descriptor, namely histogram of edge segment curvatures (HESC) whichextracts edge segments of an input image and construct a histogram from quantized curvature values of them. Therefore, HESCunveils geometric information of texture images by utilizing curve strengths for each pixel along the edge segments. We showthat the proposed feature descriptor is robust against rotation and translation. We also extend HESC descriptor to emphasis thecontribution of small curvature values. We carry out several experiments in UIUC texture dataset and compare the performanceof the proposed HESC descriptor to well-known Local Binary Pattern (LBP). The proposed texture descriptor outperformsLBP in terms of recognition accuracy