A Novel Hybrid Apporach Based on Normalized Color Spaces and 2DPCA for Color Face Recognition

Shangyou Chen,Yanfeng Sun,Baocai Yin,Caifang Song

DOI: https://doi.org/10.1109/cisp.2010.5647938

2010-01-01

Abstract:Recently, in a task of face recognition, some researchers proposed a novel method called color space normalization (CSN) which can enhance the discriminating power of color spaces for color face recognition. This method can significantly reduce the correlation between the three color components RGB and thus can enhance the discriminating power of the concatenated color component RGB. This offers an intrinsic (or internal) reason for why the CSN techniques can improve the face recognition performance. The three color component images of normalized color spaces are concatenated into one pattern vector, then PCA+FLD is performed on the concatenated pattern vector. But the concatenated pattern vector based on PCA destroyed the overall structure of each pixel in image. Some local structure information may be ignored. While 2DPCA preserves the overall structure information well and eliminates the small sample problem on a certain extent. Therefore, in this paper, color space normalization was combined with 2DPCA for color face recognition. This CSN based 2DPCA was performed and compared with PCA, 2DPCA for color face recognition. Numerous experiments show that the proposed approach has better recognition performance than other traditional PCA-based methods for color face recognition.