Research And Implementation Of Constrained Algorithm For Texture Synthesis From Multi-samples

With the development of technology of Computer virtual reality and Image,Texture Synthesis from Samples has become an important one of it. In recent years, itnot only improves texture seams and texture distortions which Texture Mappingcauses, but also avoids trivial problem of adjusting the parameters. Therefore, it hasbecome one hotspot of Computer Graphics, Vision and Image Processing. On thisbasis, this paper presents some new ideas, designs and implements the relevantalgorithms, and gives a detailed analysis process and experimental results. Specificwork includes the following:Firstly, this paper presents some basic concepts and development correlated withtexture, and we are more details to present several classical algorithms of TextureSynthesis from Samples, implements, tests and analyses these algorithms, summarizestheir advantages and disadvantages, and proposes some improved measures.Secondly, we introduces SSIM which is used for image recognition to evaluatestability and partial of image, and present a method for choosing neighborhoodwindow of texture synthesis based on structure. This method can avoid improperchoosing neighborhood window of traditional algorithms, which leads to expend timeincreasing or quality of results descending.Thirdly, we research and analyze the principle of the traditional texture synthesisalgorithm, discover some vision limitations in it, and present An Improved naturalTexture Synthesis Algorithm. The algorithm’s core includes a kind of candidate set,new error scale formula, new search path and dynamic weighted neighborhoodwindow strategy, presents a kind of improved natural texture synthesis. Experimentsindicate that the improved algorithm can effectively improve the texture synthesisresults.Fourthly, we present a new method of Texture Design and Synthesis Based onConstrained Multiple Sources. Adopting characteristics of the method of patch-basedcombining with pixel-matching texture synthesis to different areas of constrainedsample. At the same time, adopting dynamic neighborhood window. Correlativeexperiments draw a conclusion that our algorithm can solve the problems of basictexture element fragment and repetition which the others can’t, and increased synthesis rate.Our algorithms and tactics can analyze, design and process all kinds of objectscontain image, such as picture, remote sensing images, video and so on, and it can bewidely used in image beautify, learning and creating of image style and remotesensing images.