Hybrid Optimization based DBN for Face Recognition using Low-Resolution Images

Abstract

The recognition of faces has gained immense interest in image processing. The conventional face recognition techniques provide improved performance using the frontal images with high resolution. However, the major problem in face recognition is the Low-Resolution face images. To address this challenge, this paper proposes the face recognition system by integrating the Gabor Filter + Wavelet + Texture (GWTM) operator and the Deep Belief Network (DBN) to increase the classification performance, while deploying the low-resolution images. Initially, the input image is subjected to the preprocessing, and the low-resolution image is generated. Then, these low-resolution images employed kernel regression model for generating an image with high-resolution. Then, both the low-resolution and the high-resolution images are applied to the GWTM operator for extracting significant features. The result of the GWTM is provided to the fractional Bat algorithm for producing the intermediary images. Finally, the intermediary images are given to the DBN classifier for optimal face detection. The proposed method is analyzed with the existing methods using three evaluation measures, like the false acceptance rate (FAR), accuracy, and false rejection rate (FRR). Thus, the proposed method outperformed other methods with higher accuracy of 0.98, minimum FAR and FRR of 0.05.