Color transfer method based on saliency features for color images

Abstract

With growing demands for higher image quality in the fields of film, video post-production, image restoration, art creation, and computer vision, color transfer between images has become an important research area. Based on previous research on color transfer techniques, this paper proposes a color transfer method for images based on saliency features, aiming at automatic color migration between them. Transferring colors based on the saliency features of the input image can avoid the problem of unnatural color of the output image due to mixing of colors from different regions. First, the local variances of both the original and reference images are calculated, serving as a temporary saliency feature map. This is followed by obtaining a refined saliency feature map after undergoing processes such as minimization filtering, binarization, expansion, and iteration. Subsequently, color is transferred between the saliency and non-saliency regions of the original and reference images. To avoid the generation of pseudo-contours, the image is then refined using base projection. Finally, an output image is obtained by fusing the base-projected image with the outcome from Reinhard’s method, ensuring the output retains its naturalness and consistency. We conducted experiments with different types of images such as natural landscapes, buildings, and art paintings. The experimental results show that the method proposed in this paper not only retains the intricacies of the original image but also offers fuller and more realistic color renditions.