Improved face recognition method using SVM-MRF with KTBD based KCM segmentation approach

Digital image processing is a technique for visually analyzing images that can be manipulated using different approaches. It can also be used to improve image quality and group images by identifying different special features in the images. When it comes to face recognition, evaluating face images with different postures is challenging. As a result, different researchers use edge detection approaches to analyze the images, however, the classification accuracy is low due to the high computational complexity. In the meantime, assessing images with a lot of noise or poor contrast is necessary. As a result, we have presented a novel face recognition approach that uses Contrast limited adaptive histogram equalization (CLAHE) to preprocess the images and a kernelized total Bregman divergence-based K-Means Clustering algorithm to improve segmentation even for images with high noise levels. Using dominant color structure descriptors, SIFT descriptors, improved center-symmetric local binary patterns (ICSLBP), and histograms of gradients(HOG), the features are derived. After that, a support vector machine (SVM) with a modified random forest (MRF) model is used to do facial recognition. The results demonstrated that the proposed SVM-MRF method achieved better classification accuracy with low computational complexity. The main reason is due to the novel segmentation technique used. The experiments are conducted on AR, CMU-PIE, and YALE datasets and obtained better segmentation and classification accuracy than other approaches, and the feature extraction time is also reduced to a great extent.