关键词: 颜色转移, 邻域优先搜索, 纹理匹配

Abstract: To address the problem of grayscale image colorization in the fields of film and television, medical imaging, animation, space exploration and so on, a color transferring algorithm based on neighborhood-first search and texture similarity matching is proposed. For a pixel to be colorized in target grayscale image, the method first finds its candidate matching pixels based on the neighborhood-first searching scheme. Then, we define the concept of the texture eigenvector to describe the texture similarity between two pixels, and determine the final matching pixel by calculating the matching error based on the distance between the eigenvectors of the candidate pixels and the target pixel. To address the problem of image distortion occurred after global matching, the present method conducts one or more swatch matching procedures for the distortion regions of the colorized image. In the five test cases of image colorization, the proposed algorithm has a reasonable color transfer effect. Moreover, for the image with complex color composition or semantic content, this paper verifies the effectiveness of swatch matching through a case, and analyzes the problems and shortcomings of swatch matching.

Key words: color transferring, neighborhood-first search, texture similarity matching