Color Transfer Based on Visual Saliency

Abstract

Color transfer is usually used to match the color between input images and reference images to change the global color performance or local color of some specified object. The technology had been widely used in digital media, photography industry, culture and art, remote sensing, biomedicine and other relative fields. Automatically construction of the color mapping relationship between the reference image and the input image had become a key problem of color transfer. The paper proposed a color transfer intention construction model with the subjective perception prior to predicate background and saliency regions based on the Simple Linear Iterative Clustering (SLIC) color clustering of input image and reference image. In the model, a novel visual saliency algorithm is presented. Firstly, a boundary assumption is introduced to compare the contrast between different color clustering and boundaries. Then, the chromaticity distance of color clustering is defined. Finally, visual attention model is introduced to simulate the visual perception process to generate a feature map with different visual saliency factors to construct a color transfer intention based on subjective prior. 20 images are selected from ASD-1000 dataset as experiment samples to verify the model proposed in the paper. A new image quality evaluation method based on visual attention model is suggested to compare the proposed model and typical algorithms. The experimental results demonstrate that the method proposed in this paper can construct the color transfer intention effectively and obtain reasonable color transfer results.