Binary face image recognition using logistic regression and neural network

Abstract

A face recognition technique based on binary image using logistic regression and neural network is proposed. In the proposed technique, a color image is converted to gray image and then denoised using low pass filter. Local window standard deviation is applied to the denoised image to capture the local intensity variations around eye brows, eyelids, nose and mouth, which is then binarized using adaptive thresholding to get good quality binary image. Probable face region is then detected from binary image. The size of the pre-processed binary face image is then normalized and reduced to 50%, 30%, 20% and 10% of its original size using nearest neighbor interpolation method and a face database is created corresponding to each reduced size image. Reduction in image size minimizes the computational space and time for training with logistic regression and neural network. Application of logistic regression gives a training accuracy and testing accuracy of 100% while 3-Layer Back-propagation neural network (BPNN) gives a training accuracy of 100% and testing accuracy of 97.5% even after the size is reduced to 20% of the original image dimension.