Research On Image Coloring And Compression Based On LapRLS

Image coloring is the process of adding color to gray image,which needs to be completed by computer.Image coloring can also be regarded as image restoration.So far,it has been widely used in the fields of medicine,archaeology and industry.For example,in the process of precious old photos under special condition,color image has great research value because of its outstanding details and better visual effects.With the development of science and technology,there have the growing demand in the quality and the diversity of the digital image and video,the amount of data has gradually increased.For example,a 1024*512 size of the true color image,the space has reached 12 M bytes,video is up to hundreds of megabytes of the transmission.Therefore,the network capacity is also more and more high,which resulting in huge transmission pressure is also needed to be payed attention.Starting from the image,it will greatly reduce the network pressure which has reasonable compression processing and reduce the space.So it is necessary to study a good image compression algorithm.Based on the above background,this paper studies the prediction of color pixels in image compression and coloring.In the context of machine learning,image compression and image colorization can be regarded as the prediction of color points.Because of the particularity of machine learning algorithm,image compression can only store grayscale image and some representative pixels at the early stage;later,the stored information is used to model the gray image,so the image color and compression are complementary.The main idea of this paper is the Laplasse regularized least squares method in machine learning(Laplacian Regularized Least Squares,LapRLS).In the framework of supervised learning,this paper proposes a predictive function based on supervised learning and other assumptions(manifold assumption)or the internal structure and characteristics of the image.Finally,based on the kernel based nonlinear mapping method,the original data points are mapped to the low dimensional feature space to reduce the computational complexity.The research work and innovation of this paper are as follows:(1)Establish the colorization prediction model based on supervised learning LapRLS algorithm.First,establish a loss function based on the training point,using all the marked points.In order to reflect the internal structure of the data and to guarantee the stability of the final results,two regularization terms are added to the loss function.To reflect the correlation between the pixels of the image,this paper uses the manifold hypothesis.On the premise of this hypothesis,we construct the--K(K-Nearest,KNN)graph of the similarity between pixels,and the weights represent the similarity degree.The regularization term is added to guarantee the convergence of the solution without divergence,constrainting the norm of the predictive function on the basis of Gauss.Based on this model,the optimal solution is obtained.(2)Optimize the model on the problem of computational complexity on coloring problems.KPCA is used to control the computational complexity in the appropriate range using the kernel approach in machine learning.On the basis of the original data points,the non-linear mapping is used to map the original data points into a point in the feature space,and the effect is equivalent to dimensionality reduction.The kernel-based method is added to the optimization model to recalculate the coefficient vector and the optimal solution to improve the efficiency of coloring.(3)Using the LapRLS algorithm based on supervised learning establish the color point prediction and optimization model.In this paper,we study the lossy image compression,and the optimization model is established before.But it is worth noting that this paper does not use the previous learning or other heuristic learning method to select the points,therefore,the time consuming problem is avoided.In the paper,the uniform sampling method is used to provide the training points,and the optimal solution is obtained by training model.(4)In the study of image compression,in order to improve the quality of the image at the decompression phase,this paper uses the difference picture to optimize the model.By calculating the difference between the original image and the output image to store the difference image,therefore enhance the quality of the restored image.