Performance comparison of color spaces in Thepade's transform error vector rotation algorithms of vector quantization for grayscale image colorization with Slant and Hartley transforms

Abstract

The paper presents an effective monochrome image colorization technique using Thepade's transform error vector rotation algorithms with Slant and Hartley transforms in RGB, YIQ and YCbCr color spaces. Vector quantization is used to generate a color pallet which colors a target monochrome image. Two error vector sequences (Slant & Hartley) represented by binary numbers based on Slant and Hartley transforms are used in Thepade's transform error vector rotation algorithms. The proposed method works in two stages. In first stage color pallet is generated using the source (color) image from which color traits need to be taken using Thepade's transform error vector rotation algorithms and then colors are transferred to a target (monochrome) image using generated color pallet. The quality of colorization depends on the source color image chosen and the target monochrome image which is to be colored. As there are no objective criteria for qualitative analysis of the proposed method, here the grayscale version of original color image is recolored using proposed technique and as quality comparison criteria, the mean squared error between original color image and recolored image is calculated. Five different codebook (color pallet) sizes give variations of proposed colorization method. Using recolorization of 15 different images, The experimental results show that higher color pallet sizes give better colorization in RGB and YIQ color space and lower color pallet sizes give better colorization in YCbCr color space.