Performance assessment of assorted similarity measures in gray image colorization using LBG vector quantization algorithm

Image colorization is method of adding colors to a graytarget image from multichrome source image. The proposed method performs automatic colorization using LBG codebook generation algorithm with assorted similarity measures for mapping of grayimage pixels with relatively analogous multichrome image pixels. The detailed performance assessment of the different similarity measures on the quality of colorization is done here. Experimentation is done with a test bed having 28 images. The performance comparison of the similarity measures have shown that the Canberra distance and Manhattan distance out performs other considered similarity measures for LBG based colorization technique. Higher codebook sizes give better colorization quality. Hamming distance and Wave Hedges distance are not found suitable for gray image colorization.