Color Reconstruction Of Greyscale Image Based On Sparse Representation

Color reconstruction of greyscale image,which is one of the most active,challenging important research topic in image processing and pattern recognition,is attractive recent years,such as the restoration of old black and white photos,the colorization of historical pictures,health care,space exploration and so on.Therefore,there exists great practical significance to study the color reconstruction of greyscale image.With the development of compression sensing,current researchers have paid their attention on color reconstruction based on sparse representation because of its simplicity,quickness,high efficiency,reality and full automatic.A lot of methods have been proposed,however,those methods have some limitations.First,a single dictionary may result in wrong colorization because it cannot adapt to different image patches and contain different content.Second,the nonlocal self-similarity is ignored and that will lead to a certain degree of artificial blocking effect.Third,the neglected reconstruction error can cause the unsatisfactory result.Based on the theory of sparse representation,this thesis analysed the basic methods and algorithms of color reconstruction.In order to improve the reconstruction results,this thesis improves two different color reconstruction algorithms.The main contribution of this thesis is given as follows:1.An image colorization algorithm based on classified image patches and non-local sparse coding is proposed.To cope with the unsatisfactory reconstructed image caused by the low applicability of a single dictionary,the the image patches are classified by variance and entropy.Furthermore,aming at the problem of artificial blocking effect,the thesis leverages the nonlocal sparse coding to improve the sparse coeffcients,thus,the reconstructed results are improved and artificial blocking effect is avoided.2.A color reconstruction method based on K-means classification and residual compensation is proposed.To solve the unsatisfactory color problem caused by the difference between different patches of image,this thesis adaptively classify the the reference image and original greyscale image patches into K classes by using the K-means algorithm and the method of minimum centroid distance respectively.In addition,the residual compensation is applied to correct the reconstruct results so that the reconstructed color image can statisfy the requirement of the subjective vision.