Research on Portrait Matching Technology based on Feature Points

Ye Wang, Shanyou Yang, Long Cheng*

Development & Service Center for Science &Technology Talents, The Ministry of Science &Technology (MoST), Bei-jing, 100045, China

Abstract: Most of the traditional matching techniques are gray portrait matching. As a result, a new character portrait matching technique based on character attribute points is proposed. Using the SIFT algorithm, the character attribute information of the person's name, gender, place of origin, position, technical field, etc. is taken as the local feature point of the character, and the character feature data set is prepared. The feature search strategy with kd tree as the core is selected to establish the search space and search feature set. Within the feature point mapping, the similarity measure is used to judge the similarity between the feature points of the feature set through the cost function, and the character portrait matching is realized. Experimental research shows that compared with the traditional gray portrait matching method, the feature point based human portrait matching technology can improve the accuracy of human portrait matching in normal environment by 17%. In the interference environment, the matching accuracy can be improved by 25%.

Keywords: Character portrait; Matching degree; Feature points; Character attribute information