Convolutional neural network models for deep face recognition on limitation and interfering factors in image dataset

Abstract

Face recognition is one of effective method often used for personal identification, the accuracy of the face recognition depends on many factors typically implemented at different places in unconstrained environments. Not only, the amount of images in the dataset are affected to the accuracy of face recognition but also the quality of the images is also an impact. For this reason, this work proposed the convolutional neural networks model to improve the accuracy of the face recognition under an insufficient a number of images in dataset and the images that contains an interfering factors. The challenge of this work is the regulating and configuring of many parameters the network for its best performance and suite for this conditions. The experiment results shown that the CNN model gives encouraging accuracy of the face recognition. Furthermore, this work also compared the accuracy with the different face recognition techniques such as Fisherfaces, Eigenfaces, LBPH, and MLP neural networks. For these result, CNNs were used as an efficient solution for improving the rate of recognition accuracy on this conditions.