Efficient compressed Face Recognition Method using Face Detection and ANN classifier

ABSTRACT The face recognition is the challenging task to identify from various facial expressions, background illuminations and pose variations. In this paper, we propose efficient compressed face recognition method using face detection and ANN classifier. The benchmarked face databases viz., ORL, JAFFE, YALE and Essex Faces94 are used to test the algorithm. The eight-bit pixels of every face image are segmented into two groups such as Higher Order Bits (HOBs) and Lower Order Bits (LOBs) to represent significant and insignificant decimal values of a pixel. The 4-bit HOBs are considered for further processing by omitting 4-bit LOBs to reduce 8-bit pixels into 4-bit pixels which reduces complexity and memory requirement. Another advantage of only HOB with 4 bits is the number of intensity levels are only 16 inplace of 256 intensity levels in the case of 8- bit pixels, leads to increase in speed of computation. The face image part is detected using a Viola-Jones algorithm by discording unwanted portion in the face image database. The Discrete Wavelet Transform (DWT) is applied on face detected image to convert spatial domain into frequency domain with Low Pass filters (LPF) and High Pass Filters (HPF). The LPF coefficients represented by LL subband are considered by discording high frequency bands to enhance the quality of an image and reduce the dimensionality of the original image. The GIST concept is applied on LL subband to obtain compressed features of 640 coefficients. The convolution is used to fuse GIST and LL subband coefficients to obtain final features. The Artificial Neural Network (ANN) is used as a classifier to identify a person effectively. It is witnessed that the Percentage Recognition Rate (PRR) of the proposed method is better compared to the existing methods. Keywords: Biometrics, Face recognition, Face detection, DWT, GIST Descriptor, Viola-Jones algorithm