Face Recognition: A Comparative Approach from Traditional to Recent Trends

Face recognition, an important biometric method used extensively by researchers, has become more popular recently due to development of mobile applications and frequent usages of facial images in social media. A major development is attained in facial recognition methods due to the emergence of deep learning methods. As a result, the performance of face recognition systems reached a matured state. The objectives of this research are to improve the accuracy rate of both traditional and modern methods of face recognition system under illumination variation by applying various preprocessing techniques. In the proposed face recognition approach, various preprocessing methods like SQI, HE, LTISN, GIC and DoG are applied to the Local Binary Pattern (LBP) feature extraction method and by using the Weighted Entropy based method to fuse the output of classifiers on FERET database, we have shown improvement in recognition accuracy of as high as 88.2 % can be obtained after applying DoG . In a recently used approach, deep CNN model is suggested. The Experiments are conducted in Extended Yale B and FERET Database. The suggested model provides good accuracy rates. To improve the accuracy rates further, preprocessing methods like SQI, HE, LTISN, GIC and DoG are applied to both the models. As a result, higher accuracy rates are achieved in deep CNN model both in Extended Yale B Database and FERET Database. Extended Yale B Database provides the highest accuracy rate of 99.8% after the application of SQI and an accuracy rate of 99.7% is achieved by applying HE