A comparative study of face recognition algorithms on R1 face database

Bai Limin,Jia Mingxing,Qiao Shengyang,Wu Qiang
DOI: https://doi.org/10.1109/CCDC.2014.6852224
2014-01-01
Abstract:This paper introduces the development process of face recognition and analyses representative algorithms for each period. Considering different races and numbers of samples in database, and a wide variety of pose, shelter, illumination, and expressions, different algorithms are tested based on the application requirement. It adopts CAS-PEAL-R1 face database, composed entirely by Asian faces, while the previous face recognition test are almost all based on Europe and America face database. The main work is to get two indices (recognition rate and recognition time) when applying different algorithms on R1 face database and then analysis the advantages as well as disadvantages of each algorithm. According to the comparison of the indices for each algorithm, it showed that LBP algorithm achieves state-of-the-art performance in both recognition rate and time, so it meets the requirements for real-time recognition. In addition, although the SFD (Improved SIFT Algorithm) obtained the highest recognition rate in the comparison, it doesn't satisfy the requirements in real-time recognition system for its long recognition time. Contrast previous face recognition algorithms utilized on R1 face database, some more comprehensive algorithms are introduced and tested on R1 in this paper and it sure can gives a more comprehensive reference for later researchers.
What problem does this paper attempt to address?