Research And Implementation Of Image Colorization Technology Based On Local Color Transfer

Colorization of grayscale images has always been the focus of research in the field of image processing.The grayscale images contain too little information to be studied and used in depth,but color images have more prominent details.Their contents are more true,their visual effects are more in line with the ornamental,and their application scenes are wider.Therefore,colorization technology is of great importance in research and application.At present,image colorization technology has been widely used in film and television photography,animation production,medical treatment,space exploration and other fields.Colorization methods can be divided into two main categories:the first is the color diffusion method based on artificial color lines,and the second is the color transfer method based on the reference color image.Artificial line based method requires a lot of manual participation,though it works well,but it is not practical.Therefore,this paper focuses on the study of color transfer method.The key of this method is to find the appropriate reference color image and transferring color.Based on the existing colorization algorithm,two different colorization methods are proposed and implemented.The main contents are as follows:1)An image colorization method based on improved simple linear iterative clustering(SLIC)superpixel extraction algorithm is proposed.This method not only improves the defect of SLIC superpixel extraction algorithm in the process of colorization,and improves the segmentation quality at the edge of the image,but also the strategy of cascaded image matching using multiple features is proposed.Finally,color value is propagated based on local color transfer algorithm.The experimental results show that the improved scheme can effectively improve the quality of color image,and the color algorithm in this paper can achieve better color effect than other algorithms.2)An image colorization framework based on random forests and joint bilateral filtering is proposed.We use the random forests to train the superpixels extracted from the color image,and use the trained forests to predict the color information of the super pixels on the grayscale image.Then,the joint bilateral filtering algorithm is applied to deal with the correction and optimization of color values in the image.Through contrast experiments,it can be concluded that this method can achieve very good colorization results.3)A simple and efficient color system of gray image is designed and implemented.In this system,the users can choose the two kinds of color algorithms to carry out the color operation of the image,and can also adjust the color effect by setting the parameters.