Neighborhood Similarity-Based Color Transfer Algorithm

Color transfer produces a new image and keeps the content of source image and the color of reference image. The technology is widely used in areas such as prepress image processing and multimedia post-production. The article proposes an adaptive local color transfer algorithm based on neighborhood similarity. Firstly, two classical algorithms-Reinhard algorithm and Welsh algorithm-are analyzed on the applicability and deficiency according to the experiments. Then, a new method is introduced to improve Welsh algorithm on incomplete point neighborhood description and the low efficiency of searching for matching points, which is calculating the neighborhood average value by adding high-frequency texture information to describe neighborhood features, and local hierarchical matching points searching strategy. The matching points searching strategy includes dividing the target image into blocks, searching for matching point based on neighborhood similarity firstly, and searching the global image to find best matching points finally. Experiments demonstrate that the improved algorithm can transfer the referential color effectively without any user intervention.