Natural Grayscale Image Colorization via Local Sparse Coding

Colorization is a process of adding color to grayscale image under certain constraints.In this paper we present a new colorization approach via local sparse coding.Our method is based on the positive correlation between grayscale texture and color.We firstly train a joint dictionary from the reference color image,and then solve the coefficients under the locality constraints both in texture subspace and neighboring pixels.To preserve color consistency in local region,we refine the colorized result by image pyramid.Experimental results demonstrate that the proposed method outperforms the previous by providing better reality and smoothness in color field.