A Novel Method for Face Recognition Based on Features and Gestures

Abstract

This work proposes a novel method to recognize the face of an individual by extracting features from a poorly illuminated frontal facial image. The accuracy of the face recognition system is highly dependent on robust feature extraction. For this purpose, a pre-trained deep neural network model ‘FaceNet NN4’ with a minimum set of weights and fewer number of layers has been used to extract features from a given input image. The model is optimized by using triplet loss function and 128-dimensional encodings for each image have been computed to represent facial features. The distance matrix is generated by calculating the differences between encodings using standard L2 distance formulae. Based on the distance matrix, the relative similarity of image pairs taken into consideration for the recognition task. Reducing weights and number of layers have greatly contributed to the model’s effectiveness in terms of time while achieving a good accuracy rate. All the experiments are carried out on the YaleFace dataset. Experimental evaluation and analysis exhibit that the proposed method acquires a better recognition rate with improved efficiency.