A Closed-form Algorithm to Shadow Segmentation using a Single Image

Abstract—The presence of shadow in natural scenes continue to pose a challenge to computer vision algorithms. While there have been many proposed approaches to detect and segment shadow in natural scenes, these methods fail to do so in real-time without requiring a priori information. This paper presents a model that can decompose an image into the product of an illumination component L and a reflectance component R and use the reflectance component to segment the shadow region, all in real-time. This is done by exploiting the fact that light intensity has less effect on the reflectance of an image and by assuming there is a single point light source. Our method is highly effective in detecting shadows in an image with a surface material that exhibits both specular and diffuse reflectance properties.