Gabor face recognition by multi-channel classifier fusion of supervised kernel manifold learning

Motivated by the multi-channel nature of the Gabor feature representation and the success of the multiple classifier fusion, and meanwhile, to avoid careful selection of parameters for the manifold learning, we propose a face recognition framework under the multi-channel fusion strategy. The Gabor wavelet endows the algorithm in a similar way as the human visual system, to represent face features. To solve the curse of dimensionality due to multi-channel Gabor feature, as well as to preserve nonlinear labeled intrinsic structure of the sample points, the manifold learning is applied to model the nonlinear labeled intrinsic structure. Each of the filtered multi-channel Gabor features, is treated as an independent channel. Classification is performed in each channel by the component classifier and the final result is obtained using the decision fusion strategy. The experiments on three face datasets show effective and encouraging recognition accuracy compared with other existing methods.

► We regard each channel of Gabor features as an independent sample.
► Manifold learning is applied in each Gabor channel to extract sub-manifolds.
► There is no further need to specify the neighborhood parameters for manifold learning.
► Classifiers fusion strategy helps it to be robust to parameter variations.