Image Decolorization And Fusion Based On TV-Stokes Model

Image Decolorization is an algorithm to convert color images into grayscale images,which is the basis of image processing and machine vision,and the basic tool of digital printing,photo rendering and single-channel image processing.Therefore,target recognition,target detection and image retrieval algorithms need to be based on the gray image after color removal.The key problem in the study of image decolorization is how to retain the effective information of the original image to the maximum extent so that the original chroma,brightness and color saturation of the color image can be fully reflected in the grayscale image.To solve this problem,we have done the following work:First of all,based on TV-Stokes model,we extract the smallest feature vector from the first basic differential form.Our idea is to use TV-Stokes model framework and reconstruct the thought from the image restoration and image inpainting to the image decolorization.We propose a two-step model based on a TV minimization in each step and the use of geometric information of the image.Then,two numerical methods are shown.One is to use the gradient descent flow method and the finite difference scheme for numerical calculation.The other is based on the idea of optimization,Chambolle's original dual algorithm framework is presented numerically.Secondly,in order to avoid increasing the noise of the original image in the process of color removal,inspired by TV-stokes model,a gradient term containing the original image is added on the basis of Jin's model to make its gradient field match the original image's gradient field,so a new local variational model is obtained.Then,the existence and uniqueness of model solution are proved.Then,the minimization problem is transformed into a sub-problem,and each sub-problem is iterated in the alternating direction of minimization.The gradient descent flow method is used to solve the subproblem,and the finite difference method is used for implementation.Finally,comparing our models with classical method(rgb2gray),Jin's model,experimental results are reported to demonstrate the effectiveness of the proposed method,and its performance is better than those of the other testing methods,especially for the image with noise.At the same time,it also can have the effect of denoising during image decolorization,which shows a certain superiority in visual effect.