Voting-based separation of diffuse and specular pixels - Electronics Letters

Introduction: Most algorithms in computer vision assume that the surfaces in a scene are perfect Lambertian. However, the presence of specular reflection is inevitable in most images. Although many methods have been proposed to separate specular pixels from an image, they have some critical limitations. Some can only deal with uniform coloured objects [1] and some need prior information about the scene illumination and assume a single illumination source [2, 3]. We propose a fast and efficient algorithm to separate diffuse and specular pixels using a single image, which is taken under a general situation, without any prior knowledge or explicit colour segmentation.