Recognition of Face Images by DWT, NB, SVM, FFNN and CFNN Methodology

The paper proposes a conceptual methodology for face recognition, combining phases of extraction of informative features with Discrete Wavelet Transform and identification of images after decomposition with Machine Learning and Artificial Intelligence. The Haar, Coiflets, Biorthogonal and Symplets Wavelet Transforms from 2 order were integrated as tools for feature extraction of Approximation and Detail wavelet coefficients. Analytical methods based on Naive Bayes and Support Vector Machine techniques were adapted for manipulations with two personalization classes. An advantage was found when using the Nu-Support Vector Machine approach with different Kernel functions for face recognition procedures. Training was conducted with Gradient techniques in the synthesis of three-layer Feed-Forward Neural Networks and Cascade-Forward Neural Networks in the diagnosis of five face images. Unsuitability of Feed-Forward networks was found when applying Log-sigmoid transfer function in the output layers of the networks. High levels of recognition correctness were obtained with minimization of accepted error criteria when using Cascade-forward and Feed-forward neural networks in Linear and Hyperbolic tangent sigmoid activation of outputs.