A novel triangular DCT feature extraction for enhanced face recognition

The sequence of operations for face recognition comprises of the pre-processing techniques, feature extraction, selection and classification for precise identification and substantiation of subject images. Initially, the face images are processed by Gaussian of Laplacian blurring with Median and Weiner filters for the removal of undesirable noise and frequencies, a unique approach in advancing the preprocessing techniques. The pre-processed images are applied to feature extraction transformations, namely, the Discrete Wavelet Transform (DWT) coupled with Slope-form Triangular Discrete Cosine Transform (STDCT) to generate critical essential features from the images. Primarily the feature vector space is searched for optimal selection of feature subset utilizing the Binary Particle Swarm Optimization (BPSO) search algorithm based on mutual behavior of bird flocking or fish schooling. Evaluation of the subset by the Euclidean Classifier produces reliable face recognition rate. The system dependability is attained by processing with standard databases like, the Color Facial Recognition Technology (FERET), Olivetti Research Laboratory (ORL) and Japanese Female Facial Expression (JAFFE). Induction of MATLAB to analyze the imaging methodologies with standard databases, demonstrates computation of slope of the hypotenuse of a right triangle in STDCT for reduced feature extraction. This novel technology transcends other systems by generating enhanced face recognition rate with optimum selected features to achieve on execution with multiple iterations, establishing propriety and reliability of the proposed system during the process of validation.