The Research On High Precise Decolorization Algorithm

A color-to-gray conversion is the process of converting color image into grayscale image,which is a basic tool for digital printing,photo rendering and single-channel image processing.Since the color image contains three color channels(red,green and blue),grayscale image can be regarded as the conversion of three-dimensional channel information into one-dimensional grayscale data.However,the loss of color image information cannot be avoided in the conversion process.The main task of decolorization is to retain as much information as possible from the original color image and generate grayscale images according to the human eye perception for different applications.Decolorization is a basic subject in the field of image processing,computer vision and the important premise,in order to maintain color reproduction in the process of gray level change original intention and prominent features,while make the gray image conforms to the human eye perception,in this paper,we propose three decolorization algorithms that combine the points of image fusion,structural similarity of image and the spatial correlation of image,respectively:1)A color-to-gray conversion based on multiscale image fusion.The color image is decomposed into R,G,B three monochrome channel images.After applying the Laplacian pyramid decomposition,fusing the corresponding layers of each Laplacian pyramid of monochrome channel image,the coefficient matrix is determined by the contrast,saturation and saliency of each channel image.Therefore,the decolorization problem is converted to the task that preserving the three monochrome channel images' characteristics as much as possible during the process of multi-scale fusion.The algorithm is sensitive to the edge information and can preserve the features of darker or brighter areas in the color image since the Gradient Domain Guided Image Filter(GGIF)is used.2)Image structure similarity based method for decolorization.Taking the luminance,contrast and structural similarity between the color image and the grayscale image as the objective function,by discretizing a searching space,the coefficients of the mapping function are those that maximizing the structural similarity.The luminance similarity contains the information of luminance continuity and gradient information,which can effectively capture the local structure of the image and has a high sensitivity to the local structure.3)Correlation-guided decolorization using Gaussian color model.The two-stage mapping of lightness and chromaticity was adopted to transform the color image from RGB space to LMN Gaussian color space.L channel was taken as the lightness,and the correlation of each channel was solved as the chromaticity information mapping coefficient.The final grayscale image is the combination of the two normalized results.This method runs fast and the results are robust and consistent with the human eye's perception of color imagesFinally,the performance of the proposed decolorization algorithms is evaluated qualitatively and quantitatively,comparing with the existing classic and the latest decolorization algorithms.The experiments show that the proposed algorithms can maintain the contrast and main features in the original color image while the grayscale image reveal high precision.