Image Colorization Using Discriminative Textural Features

This paper presents a novel approach to scribblebased image colorization. In the work reported here we have explored how to exploit the textural information to improve this process. For every scribbled image we extract the most discriminative features using linear discriminant analysis (LDA). After that, the whole image is projected onto a discriminative textural feature space. Our main contribution lies in propagating the color in the feature space domain rather than using the luminance channel. The presented experimental validation confirms the importance of using textural information and show that our method significantly improves the obtained colorization results.