Measurement of Just Noticeable Color Difference Utilizing Free-energy Principle in Uniform Color Space

The human visual system (HVS) has a limited sensitivity in perceiving visual information such that visual masking estimation is helpful to improve the performance of image processing techniques. Most existing research efforts only focus on the methods of assessing the visual masking for gray images. In this paper, a spatial masking estimation utilizing the free-energy principle to measure just noticeable color difference (JNCD) in the uniform color space is explored for color images. According to the free-energy principle introduced recently, the HVS is sensitive to the orderly stimulus possessing structural regularity which is easily to be predicted and is insensitive to the disorderly stimulus containing structural irregularity. We reasonably deduce that the spatial masking in color images may be overestimated in the region with orderly structures and underestimated in the region with disorderly structures. Based on a simple prediction model imitating the brain works of the HVS, the structural irregularity is computed to formulate a more accurate spatial masking function of color images in the uniform color space for measuring variable just noticeable color difference (JNCD). The estimated variable just noticeable color difference is further extended to build a color visual model of estimating the visibility thresholds of color images for performance comparison. Simulation results demonstrate that the proposed spatial masking estimation for color images has better consistency with the HVS than the existing method.