An Improved Face Recognition Based on Illumination Normalization Techniques and Elastic Bunch Graph Matching
Edison Paria Fernandez, Rolando J. Cárdenas, Juan Gutierrez, Jose Galdos · 2017
Face Recognition is known to present large variability due to factors like pose, facial expression variations, changes in illumination and occlusion, among others, thus making face recognition a very challenging problem. Studies of Illumination Normalization on face images under different illumination conditions has many proposed techniques, each of them has advantages and disadvantages. The approach proposed in this paper is the integration of methods to improve quality in different illumination conditions using three different techniques like: Logarithm Transform, Histogram Equalization and Discrete Cosine Transform (DCT), applying the proposal to face recognition in situations of video vigilance, situation in which variations in illumination are one of the most decisive factors to success of face recognition, to prove the improvement offered by the proposal, it uses a method based on bio-metric features known as Elastic Bunch Graph Matching (EBGM). This proposed method had been experimented with three databases: Yale Faces A, AT&T and Georgia Tech Face Database images. Based on the results, the proposed method increases the face Recognition to 92.817% in AT&T; 98.532% in Yale Faces A and 78.933% in Georgia Database. The proposal improves the condition for different data-sets.