DFT domain Feature Extraction using Edge-based Scale Normalization for Enhanced Face Recognition

This paper proposes a novel preprocessing technique in order to improve the performance of a Face Recognition (FR) system. The proposed Edge-based Scale Normalization (ESN) process involves the use of scale normalization along with edge detection as a preprocessing technique in order to eliminate unwanted background details in face images. Feature extraction is performed on the preprocessed image using Discrete Fourier Transform (DFT). The DFT spectrums of these images extract the low frequency coefficients required for face recognition. These important features are selected through a rhombus-shaped mask around the center of the DFT spectrum. Further optimization in feature selection is achieved through Binary Particle Swarm Optimization (BPSO) technique. Experimental results, obtained by applying the proposed algorithm on Cambridge ORL and Extended YaleB face databases, show that the proposed system outperforms other FR systems. A significant increase in the recognition rate and a substantial reduction in the number of features are observed. Significant dimensionality reduction by more than 98.5% and improved recognition rate of 98% are achieved for both datasets.