Colour Constancy using K-means Clustering Algorithm

Colour cast is the ambient presence of unwanted colour in digital images due to the source illuminant while colour constancy is the ability to perceive colors of object, invariant to the colour of the source illuminant. Existing statistic based colour constancy methods use whole image pixel values for illuminant estimation. However, not every region of an image contains reliable colour information. Therefore, the presence of large uniform colour patches within the image considerably deteriorates the performance of colour constancy algorithms. This paper presents an algorithm to alleviate the biasing effect of the uniform colour patches of the colour constancy compensation techniques. It employs the k-means clustering algorithm to segment image areas according to their colour information. The Average Absolute Difference (AAD) of each colour component of the segment is calculated and used to identify and exclude segments with uniform colour information from being used for colour constancy adjustments. Experimental results were generated using three benchmark datasets and compared with the state of the art techniques. Results show the proposed technique outperforms existing techniques in the presence of the uniform colour patches and similar to Grey World method in the absent o uniform colour patches.