Development and Design of New Techniques for Face Recognition and Classification

Abstract

Face recognition is one of the most active research areas in biometric identification newline various face recognition models proposed with the acceptable performance newlineunder the supervised conditions. However, now days face recognition under uncontrolled newlinediseases such as Internet downloaded images, low-resolution images, newlinemobile, and surveillance recorded is gained significant researchers attention. The newlinesignificant variations in face illumination, pose, occlusion, and image quality are newlinekey state-of-art challenges for robust face recognition. The face recognition of plastic newlinesurgery facial images is also a challenging task. In this research work, we proposed newlinethe novel contributions towards the robust face recognition by considering newlinethe uncontrolled conditions and plastic surgery datasets. The proposed model of newlineface recognition is based on three contributions. newlineIn the first contribution, the initial face descriptor model called Hybrid Dual newlineCross Pattern (H-DCP) proposed to address the challenges of unconstrained face newlinerecognition. We used the Laplacian filter to lower the impact of illumination variations newlineand then extract H-DCP features at the component and holistic levels. In the newlinesecond contribution, we further extend the working of H-DCP to present a hybrid newlineface descriptor that bridges the gap between histogram representations and spatial newlineinformation efficiently. We applied the H-DCP and Local Directional Pattern (LDN) newlineon pre-processed face image, and then fuse its outcomes to generate the face code. newlineThe proposed face descriptor address the challenges related to variations in pose, newlineexpression, illuminations effectively, and efficiently. After the face descriptor, histogram newlinefeatures at different levels extracted. In the third contribution, after the newlinedesign of a novel hybrid face descriptor model, we focused on the design of effective newlinefeature extraction methods. The histogram features not enough to generate newlinethe most reliable and unique set of features, thus to improve the robustness of the newlineface recognition model, we design t