Image Colorization Using Linear Neighborhood Propagation And Manifold Preservation

With the development of computer technique,image colorization are transmitted and applied widely.Many researchers pay more attentions on image colorization.Image colorization technology can be widely used in many fields such as ancient painting restoration,film and television production,advertisement design,image communication,video editing and so on.It is a frontier research subject of digital image processing and application.In this paper,we research on the basic theory of image colorization and propose several image colorization methods.The contributions of this paper are listed as follows:1.Image colorization without edge crossover of colors based on anisotropic diffusionPartial Differential Equation(PDE)can be used to image colorization.By defining an adaptive diffusion tensor,a nonlinear anisotropic diffusion function based on Partial Differential Equation is established.Colors are diffused uniformly and quickly in smooth areas,while anisotropic diffusion takes place near edges,which effectively improves color transition by suppressing unwanted crossover.Numerical solution of the PDE based on a finite difference method gets satisfactory colorization effects and better quality in edge regions compared with the popular techniques.The colorization algorithm based on anisotropic diffusion is proposed to overcome the problem of edge crossover of colors.2.Image colorization using linear neighborhood propagation and weighted smoothingWe propose a colorization method based on linear neighborhood propagation and weighted smoothing.The image is initially colored by hands,and this is followed by the development of a linear neighborhood matrix model.The process is completed by global colorization using the priority method.The final colorized image emerges through a filter-based smoothing treatment from the boundary to the smooth area.Compared to currently popular colorization methods,the proposed method simplifies the requirements for the initial colorization and also produces clearer,natural,and higher-quality images.3.Image colorization method using manifold preservation and quad-tree data structuresThe method of manifold learning is applied to image colorization.The method of manifold preservation is used to establish the optimal weights.An adaptive color diffusion equation is constructed.We use an adaptive weight function built on cell corners instead of individual pixels,and because the number of corners is smaller than the number of individual pixels,this results in a runtime performance improvement.The edits of all pixels can be computed by interpolating from the edits solved from the clusters.Compared with previous approaches,our approach requires less time without sacrificing the quality of the visualization.4.Image colorization based on k nearest neighbors and locally linear embeddingWe propose to use locally linear embedding(LLE)to perform image colorization.We make innovations from two aspects.KNN is adopted to find the neighbors of the similar features with the sample pixel.We represent each pixel as a linear combination of its neighbors in feature space according to LLE.The image will be reconstructed by using the weights computed according to LLE.We have demonstrated our manifold preservation edit propagation on various applications including colorization and color replacement.The experimental results have proved the effectiveness of the proposed algorithm.The method proposed in this paper provides several effective methods for image colorization.With the development of image processing tools,image colorization technology is also constantly improving.Further improve the image colorization technology is a challenging long-term task.