Sparse Representation based Face Recognition for Occluded Faces and Unconstrained Environment

Abstract

Abstract newlineFace recognition technologies have grown in importance over the past few decades. Terrorism newlineis a problem in today s society, and the security system has to be strengthened for newlinesafety reasons. Security is a difficult job, but technologies like biometrics and security newlinecameras may improve it. Face recognition is being researched for security applications newlineby experts as the usage of surveillance cameras grows. These technologies are being newlineutilized worldwide, but a robust automated facial recognition system that can be used to newlinerecognize a person in an unconstrained environment is in demand these days. Although newlineexcellent accuracy for neutral frontal faces has been obtained, previous techniques have newlinedemonstrated poor performance for unconstrained images. Sparse representation based newlineclassification (SRC) has recently demonstrated state-of-the-art results in face recognition newlineon unconstrained face images. Over the last few decades, several researchers have newlinedeveloped extended SRC methods. This study focuses on sparse representation based newlineclassification and its extended face recognition methods. SRC and its extended methods newlinehave been addressed in terms of their limits and benefits. SRC approaches were newlineexamined based on five face recognition issues: linear variation, nonlinear variation, newlineundersampled, pose variation, and low resolution. newlineTo represent a test sample, SRC requires discriminative features. Discrete wavelet transform newline(DWT) and histogram of oriented gradients (HOG) have also been used to extract newlinediscriminative features that are robust to unconstrained images. The convolutional neural newlinenetwork (CNN) model performed better on visual input and drew more attention newlinedue to automated feature extraction. CNN is insensitive to unconstrained variation, but newlineSRC is. This study takes the advantage of the aforementioned approaches and proposes newlinedifferent SRC variants that improve the performance of unconstrained variation. newlineix newlinex newlineThe images are also influenced by occlusion in an unconstrained environment such as newlinea mask, sungla