Removal of highlights in dichromatic reflection objects using segmentation and inpainting

Abstract

In this paper, we propose a method to separate diffuse and specular reflection in dichromatic reflection objects without the need for color space analysis. Based on the histogram distribution of the input image, we assume that a specular reflection tends to increase the brightness value and thus it can be separated from diffuse reflection using thresholding technique. To find the optimal threshold values, we use the chaotic clonal selection algorithm. Then, the segmented specular highlight area is replaced with an optimal diffuse patch based on coherency sensitive hashing. The experimental results demonstrated that the proposed specular removal method has comparative performance compared to the standard single-based specular removal methods.