Research On Image Colorization And Its Application In Video

Colorization is a challenging research subject in the domain of image editing. It is the computer-assisted art of adding color to a monochrome image or movie. Colorization is widely applied in screen,medical treatment,space exploration and many industrial and agricultural technologies. The research on colorization is of great significance in many applications.This paper proposes a series of methods to improve some existing image and video colorization technologies. First, improvement on chrominance blending based method is proposed. This chrominance blending based fast coloring technology has two major defects, one is complicated user interaction, the other one is the colorization effect depending on position of color strokes. A sampling method to the recolorization of color images is introduced to improve the two limitations mentioned above. Also, the color matching method for the sampled pixels and the means of adjusting weights before blending are explored.Example image based colorization methods can provide abundant color information to make more vivid effects. It can also reduce the workload of user interaction. Aimed at the disadvantages of many example based solutions, a scheme for colorization using reference color images without user interactions is proposed. It not only improved the color transfer method by illumination remapping that requires user interaction and long processing time, but also improved the example based method by training and classifying that has complex pre-processing.Finally, the application of colorization in video processing is explored. A simple colorize solution for unaltered scenes in videos is presented. Then a patch match based correction method is proposed to improve the error accumulation effect by our basic solution. For unaltered scenes, optical flow can also be made use of to track the objects in video. For the videos with complicated details and varying scenes, the idea of using key frame extraction technology is brought up to reduce the user interactions as far as possible.