2D face recognition

An overview of selected topics in face recognition is first presented in this chapter. The BioSecure 2D-face Benchmarking Framework is also described, composed of open-source software, publicly available databases and protocols. Three methods for 2D-face recognition, exploiting multiscale analysis, are presented. The first method exploits anisotropic smoothing, combined Gabor features and Linear Discriminant Analysis (LDA). The second approach is based on subject-specific face verification via Shape-Driven Gabor Jets (SDGJ), while the third combines Scale Invariant Feature Transform (SIFT) descriptors with graph matching. Comparative results are reported within the benchmarking framework on the BANCA database (with Mc and P protocols). Experimental results on the FRGCv2 database are also reported. The results show the improvements achieved with the presented multiscale analysis methods in order to cope with mismatched enrollment and test conditions.