Unified Texture Descriptor in the Form of Color and Machine Learning, Applied to Face Identification

Facial recognition is widely employed in identification systems where access restriction is a desirable quality, providing greater security and preventing the access of unauthorized parties. With this problem in mind, this paper proposes a new texture descriptor that works in the RGB, YCbCr and HSV color modes. It obtains perceptual descriptive data from the photographs; the descriptor considers similarities between neighboring pixels and diminishing noise interference, making it robust to minor disturbances and bringing greater stability between similar images. Furthermore, the descriptor achieves good classification performance when used with both a neural network (NN) and the statistical classifier Support Vector Machine (SVM) on color images. A performance comparison of the proposed descriptor against LBP and OC_LBP employing three different face databases is presented to demonstrate that the proposed descriptor can work with the RGB, YCbCr and HSV color modes, presenting the best results when employing a neural network as a classifier with color image databases in the three studied color modes.