An Optimum 2D Color Space for Pattern Recognition

This paper presents an optimum color conversion from the 3D RGB space into a 2D selected space to the purpose of pattern recognition. The method is based on the Karhunen-Loève transform (KLT), also known as Principal Component Analysis (PCA). The resulted 2D space is defined by the two color components (called C1 and C2), corresponding to the two largest eigenvalues of the RGB pixel covariance matrix. Using the above color projection technique, we propose a color face recognition system based on feature fusion of the C1 and C2 components and a concurrent neural network classifier. The proposed system is experimented for a color face database containing 3520 color images of 151 subjects. We also present a color image segmentation using pixel clustering in the 2D color space by means of a self-organizing neural network. The new 2D color projection model may have wide applications in the areas of color-based pattern recognition.