Non-local Total Variation Model And Its Application In Image Colorization

Image colorization is the process of turn a grayscale image into a color image by computer technology.At present,it is widely used in medical treatment,space exploration,film and television,color restoration of ancient arts and so on.In the study of image colorization methods,the method based on partial differential equations and sparse representation have been one of the research hotspots.This paper focuses on image colorization methods based on coupled nonlocal total variation and sparse representation dictionary learning.The main research contents and innovations are as follows:?1?Aiming at the disadvantages of traditional TV model for image colorization,in this paper we introduces non-local operators into the TV model and propose a coupled total variation model based on nonlocal operators for image colorization.Then,we develop a fast algorithm to numerically solve the proposed model by incorporating the alternating direction method of multiplier?ADMM?,and give the convergence of the algorithm.Finally,numerical results are reported to show that this model can effectively avoid inhomogeneous color diffusion and realize fast colorization for the grayscale images with rich textures.?2?First,we apply image sparse representation theory to image colorization,and propose a new model based on sparse representation with L₀ norm for image colorization.Then,combined with KSVD?K-Singular Value Decomposition?algorithm,we give an effective algorithm to solve the proposed model by combining KSVD?K-Singular Value Decomposition?method.Finally,numerical results demonstrate that the model can effectively colorize the grayscale image and keep the color consistency of the colorized image.?3?Since solving the L₀ norm problem of sparse representation is a NP-hard problem,in this paper we further propose a relaxed model for image colorization based on dictionary learning with L₁ norm.Then,we develop an effective algorithm to numerically solve the proposed model by incorporating the L₁ convex optimization method.Finally,experimental results show that the overcomplete dictionary learned by the algorithm is highly adaptive,and the colorization results of dictionary learning are better than DCT dictionary.