A Lightness-aware Nearest Neighbor Search Method for Exampled-based Color Transfer

Transferring the color from an example image to a source image is a fundamental task in image editing. One recent color transfer method builds the connection between the source and example image under a mixture modeling framework, and utilize the EM algorithm to automatically modify the color style of the source image based on the example image. However, in real-world applications, users may want to intentionally generate multiple transferred results by varying a certain key feature in color style, such as lightness. In this paper, we propose a lightness-aware nearest neighbor search procedure that allows users to select the lightness style of the transferred image. Such a procedure selects the pixels in the example image according to the lightness to update the probability matrix in optimizing the parameters for the mixture model. Experiments demonstrate that our method can generate color transferred results with high/low lightness and high example consistency.