In this paper, a set of masks is introduced which are used in combination for simulating pattern-color separability function of the human visual system. These kernels provide a perceptual degree of smoothing, corresponding to measurements estimated from human psychophysical experiments. It has been found that these kernels yield a better approach for smoothing color images than the traditional widely practiced Gaussian masks. The masks are applied to color texture segmentation and some results are presented. The masks are also useful for other vision tasks where smoothing is a major step.