Research Of Color Transfer From Multiple Reference Images

With the development of intelligent mobile phone and mobile Internet,people can get photos and videos from anywhere at any time.Every day on the Internet there is a huge amount of image information generated,a large part of which is used to save or share,at the same time image processing technology has been widely applied in order to obtain the desired results.Due to the conditions of light,the color differences of shooting equipment or terminal equipment,the image is usually not the expected results.Color transfer is a kind of image processing technology to solve these problems effectively.Color transfer produces a composite image based on the input image and the reference image,keeping the content of the input image and the color of the reference image at the same time.Most color transfer algorithms are based on a single reference image.When there are complex scenes in the input image,it is difficult to find the appropriate color reference in one reference image for all regions in the input image.In order to solve this problem,this paper proposes a multi-reference image color transfer algorithm based on Gauss Mixture Model color clustering.Firstly,obtaining different clustering fields.Use the Gauss Mixture Model to fit the color distribution of the image that each Gauss component describes the color distribution of a region.Then use the Expectation-Maximization algorithm to estimate the parameters of Gauss Mixture Model.Gauss Mixture Model and the EM algorithm is improved from two aspects,one is adding spatial smoothing parameters to make the edge of the area smooth,the other is changing the Gauss component combination condition in order to get a more accurate Gauss component.Then,this paper presents a method of establishing the mappings between the color category regions of the reference image and the input image,so that each area of the input image can get the best matching reference color area.Then,different methods are used to complete the color transfer for grayscale images and color image.Finally,an objective quality evaluation is proposed to evaluate the color transfer results.Experiments are done,comparing with the other color transfer algorithms,to verify the rationality and effectiveness of the proposed algorithm.