Optimized Feed Forward Neural Network for an Efficient Pose Invariant Face Recognition System

Abstract

‘Face recognition’ (FR) have an imperative role in personal verification in addition to recognition. A numerous FR algorithms were developed during last decades although these approaches are much better but the accuracy of the image is less. To overcome such difficulties this paper proposed a pose invariant FR system using an optimized ‘Feed Forward NeuralNetwork’ (FFNN). This approach includes four phases explicitly a) preprocessing, b) segmentation, c) feature extraction (FE) d) classification. Primarily, the input video is transmuted into frames. Then, preprocessing is done utilizing Modified Wiener Filter (MWF). In the subsequent phase, Viola-Jones (V-J) algorithm is utilized for segmentation of the face. The nose, mouth, left and also right eye are segmented. FE is done using ‘Local Tetra Pattern’ (LTrP), SURF (‘Speeded Up Robust Features’), SIFT (‘Scale Invariant Feature Transformation’) and the DWT (Discrete Wavelet Transform). Lastly, classification is implemented using the FFNN that is optimized utilizing the ‘Adaptive Particle Swarm Optimization’ (APSO). The proposed work’s output determines if the individual is recognized or unrecognized. The experimental outputs are attained for the proposed one and they are contrasted to the existent techniques.