%0 Journal Article
%A GAO Huan
%A TANG Ling
%A WEI Shi-min
%A ZHOU Xiu-zhuang
%A CHENG Peng-fei
%T A Feature Point Descriptor for Texture-Less Images
%D 2021
%R 10.13190/j.jbupt.2021-072
%J Journal of Beijing University of Posts and Telecommunications
%P 13-19
%V 44
%N 6
%X Most of the existing feature point extraction algorithms are suitable for processing images with rich texture, but they cannot extract effective feature points for texture-less images. To solve this problem, a new feature point descriptor, termed as multi-neighborhood structure tensor features (MNSTF) is proposed. The algorithm is based on a series of fixed neighborhoods and image structure tensors. By expressing the structure and texture information of local images, it solves the problems such as feature point extraction and matching in texture-less and non-texture scenes in existing feature point extraction algorithms. At the same time, by calculating the relative direction between neighborhoods, the rotation invariance of the MNSTF feature descriptor. The experimental results show that the accuracy of feature point matching of the MNSTF algorithm on the rotated texture-less image test set is over 99.9%, which verifies the good applicability, rotation invariance and robustness of the MNSTF feature descriptor.
%U https://journal.bupt.edu.cn/EN/10.13190/j.jbupt.2021-072