Comparison of Grayscale Image Colorization Methods in Different Color Spaces

With the aim of coloring old black and white photos, medical image illustrations, and classic movies, the gray-scale image colorization methods are used to assign color information to grayscale image. Color space, as the basis of quantitative color information, plays an important role in the gray-scale image colorization. In this paper, the different color spaces—Lab, Luv, YCrCb, YIQ used in the grayscale image colorization are analyzed. Two classical automatic colorization methods, Welsh approach and Gupta approach, are carried out in those color spaces. Different performances are observed in such color spaces when using the two colorization methods. In Welsh approach, the transfer result depends on the luminance information of reference image. Since the process of Gupta approach is on the purpose of propagating color information using the least-squares optimization method, the result shows limited relevance to the reference image luminance. The experimental results demonstrate that YCrCb and YIQ have better performance in texture similarity than Lab and LUV at both color transfer methods. LUV presents the worst performance for most of the images when applying color migration. The optimal results are obtained based on YCrCb and YIQ in Welsh approach. While, it is observed that Gupta method has limited effect on the colorization results.