A Kernel Based Algorithm for Fast Color-to-Gray Processing

Color-to-gray is a classical manipulation in image processing. In this paper, a kernel based method is presented to carry out color-to-gray operation. It is noticeable that, in a meaningful color image only the linear combination of the three color components of every pixel is hardly enough to describe the implication relation between the luminance and chrominance in every pixel, or between pixels. The motivation of this approach comes from the kernel principle component analysis. The kernel method is introduced to capture and extract implicitly the nonlinear relationships indirectly. Different from the conventional Kernel PCA, this approach can perform the color conversion much faster while preserving salient features of the color image as much as possible. An optimal kernel function is designed to enhance our method's effectiveness. Experiment results shows that our method is appreciable and competitive in quality with great advantage in performance.