Illuminant Invariant Face Recognition by the Fusion of Incremental and Artificial Neural Network Approaches
Linda Sara Mathew · 2012
Dimension reduction is critical for face recognition applications as the data dimensionality is much higher than the size of the training set, leading to the small sample size problem. The issue of singular scatter matrices has been mitigated by GSVD and Incremental approach overcomes the scalability problem. In this paper, an incremental algorithm ILDA-GSVD has been used for generating the projection matrix and neural network classifier discriminates the face images.This method consists of preprocessing, dimension reduction, feature extraction and classification using neural network.Several methods are adopted to standardize the faces illumination reducing the variations for further features extraction; which are extracted using the image phase spectrum of the histogram equalized image. The incremental approach proved efficient compared to the batch method both in terms of misclassification and time even when a few samples of images are available and neural classifier further reduced the misclassification caused by not-linearly separable classes. The proposed method was tested on both feret and orl face databases. Evaluation results show that the proposed feature extraction scheme, when used together with the neural network classifier, provides a recognition rate of 92% and a verification error lower than 0.04%.