Research On Image Enhancement Using The Color Information

Image enhancement is a special type of image processing method with the purpose of prominenting some information according to the need of certain application,and weakening or eliminating some unrelated information to get more suitable images for certain application.The traditional image enhancement methods are usually conducted on the grayscale images.In comparison to the grayscale image which has a single channel,the color image with three channels has more abundant information.Processing the image based on its color information,will bring new research contents and methods to the traditional image enhancement research only to grayscale image.This thesis concentrates on exploring the color information based image enhancement,utilizes the characteristics of the multichannel color image to enhance the image and meet the demand of certain application.The thesis mainly involves three parts: one is the image enhancement method focusing on adding colors to the grayscale image which is known as image colorization,another is the image enhancement method focusing on promoting the accuracy of the reconstructed chrominance images under the color image super-resolution(SR)reconstruction framework which is kowns as chrominance image SR reconstruction,and the last is the image enhancement method focusing on preserving the features of the color image during the color-to-gray conversion which is known as image decolorization.The main contributions of this thesis are as follows.1.In the research of image colorization,in order to overcome the shortcomings of strict requirements for quantity and location of the manual color scribbles and easy to suffer color spread errors around the edge areas,this thesis proposes a colorizationalgorithm based on neighborhood similarity pixels searching and active set optimization(NSAS colorization).Using the neighborhood similarity pixels which considering both the space distance and value difference of two pixels to replace the usually adopted rectangle or rhombus window neighborhood pixels,to avoid the influence of the large difference neighbor pixels to the present pixel's local weighting relation and reduce the edge distortion of the colorization result.Translating the autoregression model of the local weighted summation formula to the quadratic programming problem with inequality constraints and linear equality constraint,and utilizing the active set optimization method to obtain the weight values,this manner can avoid the dependence on the empirical formula for weighting,enhance weight values' accuracy and bring more reliable theoretical foundation.2.In the research of chrominance image SR reconstruction,considering the luminance image has more details and edges information than that of the chrominance images,and the color-spread based colorization method using the luminance image to calculate the local weight values of the chrominance image will enhance the edge performance of the chrominance image,this thesis firstly proposes a method integrating the optimization colorization to the iterative back projection(IBP)model(IBPCSR),in order to enhance the method's processing capacity around the image edge areas without adding extra steps specially for the edges.Then,to reduce the running time of the method,this thesis proposes a method taking the luminance image as the reference and basing on optimization colorization and guided filter algorithms(GFCSR).Lastly,because the kernel function of the guide filter which takes luminance image as the guided image is conform to the local constraint of the chrominance weight values suggested by colorization research,and the procedure of IBP is conform to the global constraint of the SR reconstruction,this thesis proposes a method basing on the projection onto convex set(POCS)theory which integrates the local and global constraints to get the reconstructed chrominance images(POCSCSR).The last method can get results more similar to the original high resolution chrominance images,and the whole running time is just a little longer than the interpolation method which is generally adopted to conduct on the chrominance images in the color image SR reconstruction framework.3.In the research of image decolorization,according to the properties of human visual system and the target of maintaining visual perception consistency before and after the conversion,this thesis proposes two decolorization methods which take advantage of the image saliency feature.One is based on the global color saliency feature,which first conducts the linear weighted mapping on the three channels of the color image to get the candidate grayscale images,and then computes the similarity between the global saliency map of the color image and that of the candidate grayscale image to select the result image(PSPGD).This method can obtain the grayscale image well reservation the color contrasts of the original color image,and without brings those nonexistent abrupt edges in original image.But the cost of computing the global contrast for each pixel or color is high,this thesis further proposes the decolorization method based on the region saliency features for this problem,and combines the pixel gradient features to reduce distortions of the edges and details(PGRSPD).This method utilizes human visual properties to add chrominance items in the linear weighted mapping function so as to improve the comprehensiveness of candidate grayscale images.And adopts two-steps manner which first decreases the number of candidate grayscale images and then computes saliency maps to further bring down the calculation cost.The results of the proposed decolorization methods well remain the visual perception effects of the initial color images subjectively,and get high ratings under the measurements using two kinds of objective evaluation criterion.