提出了一种基于邻域优先搜索和纹理相似度匹配的色彩转移算法。首先设计了邻域搜索方案在彩色图像中寻找目标灰度像素的候选匹配像素, 然后基于纹理相似度计算候选像素与目标像素之间的匹配误差, 并据此确定最终匹配像素。对于不满足邻域匹配条件的目标像素, 通过对彩色参考图像进行边缘保持平滑滤波并设置采样间隔的手段, 提高其全图搜索和匹配速度。针对全局匹配后图像中出现的失真区域, 在彩色图像中设置对应的样窗与其进行单独匹配, 以进一步提升色彩转移效果。主观评价结果显示: 在4个测试案例中, 本文算法与经典Welsh算法具有相似的色彩转移效果, 但在计算效率上, 本文算法比Welsh算法具有较大优势。借助一个色彩构成较为复杂的灰度图像彩色化案例, 验证了样窗匹配在色彩转移和色域再现方面的改进效果。

A new, general, fast and user-friendly method to solve the problem of colorizing grayscale images is proposed. The neighborhood-first searching scheme is designed firstly to find the candidate matching pixels of a target grayscale pixel. Then, the final matching pixels are determined by calculating matching error between the candidates and the target pixels based on the texture similarity. For the pixels that don’t meet the neighborhood matching criterion, the whole image searching and matching can be sped up by using edge-preservation smooth filter to smooth the color reference image and setting sampling interval. To address the problem of image distortion occurred after global matching, our method conducts one or more swatch matching procedures to transfer the desired color moods from groups of swatch regions in the color reference image to the distortion regions of the colorized image. The subjective evaluation results show that in four test cases, the proposed algorithm has similar performance to the classical Welsh algorithm on the color transfer quality, but in the aspect of computing efficiency, the proposed algorithm has a great advantage over Welsh algorithm. By means of a test case of colorizing a grayscale image with complex color composition, the improvement of the swatch matching in color transfer and gamut reproduction is verified.