Face Recognition Based on Windowing Technique Using DCT, Average Covariance and Artificial Neural Network

Abstract

The field of Face Recognition (FR) is still a thought-provoking problem, while in recent advances of Artificial Neural Networks (ANN) has shown improved performance in FR rate. In this paper, we propose face recognition based on windowing technique using Discrete Cosine Transform (DCT), average covariance and ANN. The novel concept of windowing technique is used to divide each image to $\mathbf{4x4},\mathbf{8X8}$ and $\mathbf{16X16}$ size of windows. The DCT is applied on each window to obtain DCT co-efficients. The covariance matrix is computed on each DCT coefficient matrix and average value of each block is also computed to obtain final feature value. The computation of an average covariance reduces the original size of face image by around 97% i.e., the number of co-efficients in the final feature set is only around 3% of the original size of an image. The proposed method is very efficient in identifying with very less number of features. Network is created and trained the input dataset and target dataset to reach the desired output. The trained net is then tested to compute performance parameters of the network. The experiments are conducted on some popularly used face databases to illuminate the performance and the efficiency of the proposed algorithm. The experimental results are tabulated and are compared with the existing methods. It is observed that, the proposed model achieves better recognition accuracy for $\mathbf{16X16}$ windowing and also with existing algorithms.