A Feature Fusion Method for Effective Face Recognition Under Variant Illumination and Noisy Conditions

Faces extracted in bad light are affected in terms of unequal contrast, noise, and variant illumination. These kinds of disruptions decrease the accuracy of facial authentication real and complex environments. In this paper, a feature fusion method is provided to achieve illumination-robust face recognition. In this model, the Gaussian filter and Gabor filters are applied on facial image to generate the illumination-variant features. Each of the Gaussian and Gabor face is processed by LBP filter to generate the effective visual description for face. A block-level feature fusion is applied on Gaussian-LBP and Gabor-LBP faces to generate the composite feature pattern. This most relevant and adaptive feature patterns are processed on SVM classifier to recognize the face accurately. The proposed feature fusion model is applied on illumination, noise, and contrast-variant sample sets of extended Yale databases. The comparative results against SVM, KNN, and ANN methods verified the significant gain in accuracy of facial identification in complex environmental conditions.