Research On Saliency Guided Color Transfer

With the popularization of computers and the development of Internet technology,digital information dissemination has become prevalent,and images have become the most representative way of information communication and a part of most people’s lives.Therefore,as one of the key technologies of image processing,image editing has become a very important research topic.Various image editing methods have been developed,including the image adjustment,image enhancement,image collage,etc.Since color and style are important components of an image,color transfer has received a lot of attention from industry and academia in recent years.Specifically,color transfer is the process of adjusting the color of an original image based on the color of another reference image so that the original image has similar visual color characteristics as the reference image,and color transfer is often used in image or video processing,image correction,image enhancement,and other fields.In recent years,researchers have proposed many images color transfer methods,such as color transfer based on statistical information,semantic-based color transfer and user-assisted color transfer methods for doodling,and deep learning-based color transfer methods.Meanwhile many researchers propose color transfer methods by analyzing color distribution and semantic relevance.However,these methods ignore the human visual system’s sensitivity to color information in salient regions,and the color mixing situation in salient regions during color transfer affects the user’s intuition and visual experience.To solve the above problems,we design a real-time instance segmentation model for saliency differentiation of front and back view regions of images,and based on this we propose a saliency-guided color transfer depth model for independent semantic color transfer and luminance-optimized color blending of separated front and back view regions,which improves the overall color transfer effect.The specific research is divided into the following two aspects,on the one hand,a real-time instance segmentation model with improved multi-scale anchor frames is proposed.On the other hand,a luminance-optimized color transfer depth model based on saliency guidance is proposed.The experimental results show that our proposed method image color transfer is highly robust and generates better image quality performance compared to other existing studies.