Colour Constancy For Non‐Uniform Illuminant using Image Textures

Colour constancy (CC) is the ability to perceive the true colour of the scene on its image regardless of the scene’s illuminant changes. Colour constancy is a significant part of the digital image processing pipeline, more precisely, where true colour of the object is needed. Most existing CC algorithms assume a uniform illuminant across the whole scene of the image, which is not always the case. Hence, their performance is influenced by the presence of multiple light sources. This paper presents a colour constancy algorithm using image texture for uniform/non-uniformly lit scene images. The propose algorithm applies the K-means algorithm to segment the input image based on its different colour feature. Each segment’s texture is then extracted using the Entropy analysis algorithm. The colour information of the texture pixels is then used to calculate initial colour constancy adjustment factor for each segment. Finally, the colour constancy adjustment factors for each pixel within the image is determined by fusing the colour constancy of all segment regulated by the Euclidian distance of each pixel from the centre of the segments. Experimental results on both single and multiple illuminant image datasets show that the proposed algorithm outperforms the existing state of the art colour constancy algorithms, particularly when the images lit by multiple light sources.

http://icbisp2017.events.theiet.org.cn/