Gray Image Colorization using Thepade’s Transform Error Vector Rotation With Cosine, Walsh, Haar Transforms and various Similarity Measures

Abstract

The paper presents various gray image colorization methods based on vector quantization for performing automatic colorization. To colorize gray target image by extracting color pixels from source color image, Thepade’s Transform Error Vector Rotation vector quantization methods such as Thepade’s Cosine Error Vector Rotation (TCEVR), Thepade’s Walsh Error Vector Rotation (TWEVR) and Thepade’s Haar Error Vector Rotation (THEVR) are used along with varied similarity measures. The quality of colorization of gray image is subjective to the source color image and target gray image (to be colored). Here the image test bed of 25 images is used to recolor the gray equivalent of the original color images for qualitative performance comparison of proposed colorization methods with help of PSNR between original color and recolored images. Colorization is performed using diverse similarity measures which belong to different families. These nine similarity measures are used for mapping gray image pixels with relatively corresponding multichorme image pixels. When these similarity measures are assessed for their comparison for colorizing the target gray image, it is observed that Chebychev outruns all other similarity measures and the worst performance is consistently given by Jaccard and Hamming distances. Among all the considered colorization methods Thepade’s Haar Error Vector Rotation is much suitable algorithm for performing gray image colorization.
DOI: 10.17762/ijritcc2321-8169.1505166