Highlight Removal with Orthogonal Decomposition

Abstract

In this paper, based on orthogonal decomposition, we present a robust and effective method for highlight removal. First, we obtain the reflectance image of an image through orthogonal decomposition. Then, the reflectance image is used to cluster the image. According to the clustering results, illumination chromaticity is estimated. Finally, we separate diffuse and specular reflections per pixel according to the distance from each pixel chromaticity to illumination chromaticity, where the diffuse reflection image is the highlight removal image. According to our extensive experimental results, the proposed method outperforms all the existing state-of-the-art (SOTA) methods according to the Peak-Signal-to-Noise-Ratio (PSNR) and the Structural Similarity (SSIM) Index score.