Image Colorization Method Using Texture Descriptors and ISLIC Segmentation

We present a new colorization method to assign color to a grayscale image based on a reference color image using texture descriptors and Improved Simple Linear Iterative Clustering (ISLIC). Firstly, the pixels of images are classified using Support Vector Machine (SVM) according to texture descriptors, mean luminance, entropy, homogeneity, correlation, and local binary pattern (LBP) features. Then, the grayscale image and the color image are segmented into superpixels, which are obtained by ISLIC to produce more uniform and regularly shaped superpixels than those obtained by SLIC, and the classified images are further post-processed combined with superpixles for removing erroneous classifications. Thereafter, each pixel of the grayscale image is assigned with a color obtained from the color image following a predefined matching metric based on the superpixels and the classes. Experimental results show that our proposed approach is effective and has a better colorization in naturalness compared with Welsh algorithm and unimproved SLIC strategy method.