Performance Evaluation of Weighted Entropy Based Fusion Technique for Face Recognition with Different Pre-processing Techniques

Abstract

In this work, preprocessing methods are applied to both whole image and block sized image to improve the performance of face recognition. In holistic method, preprocessing methods like Locally Tuned Inverse Sine Nonlinear (LTISN), Self-Quotient Illumination (SQI), Gamma Intensity Correction (GIC), Histogram Equalization (HE), Contrast-Limited Adaptive Histogram Equalization (CLAHE) and Difference of Gaussian (DoG) filtering are applied to the complete image (holistic) and then the images are block sized. Subsequently, K- Nearest Neighbor (K-NN) classifier and weighted entropy-based fusion method are applied. In the block-based method, input image is divided into small blocks in a non-overlapping pattern. Then above-mentioned preprocessing techniques are applied to each block. Subsequently K-NN classifier and weighted entropy-based fusion method are applied to the output of each classifier. The recognition accuracy is computed for various block sizes and threshold values. This paper presents comparison of various pre-processing methods applied to both the holistic and block-based methods to mitigate the effects of illumination variation and preserve the essential details that are needed for recognition. The applied preprocessing methods enhance the image quality and increases the face recognition accuracy on FERET dataset under lighting variations. The block-based weighted entropy-based fusion method performs better than holistic method in FERET database. The maximum accuracy of the proposed method increases from 85.1% to 88.2% after the application of preprocessing methods.