Face recognition based on fuzzy integral and wavelet transform

Recently, wavelet transform and fusion technology have been used in face recognition. In this paper, we propose a new face recognition method based on probability-based fuzzy integral and wavelet transform. Firstly, an image is decomposed with wavelet transform (WT) to three levels. Secondly, the Fisherface method is applied to three low-frequency subimages respectively. Then, the individual classifiers are fused with fuzzy integral based on probability measures. Experimental results show that the new method outperforms individual classifiers. Furthermore, we compare the fuzzy integrals based on probability measure with those based on the commonly used g(lambda) - fuzzy measures with lambda > -1 and lambda not equal 0, and find that the former produces consistently better results than the latter.