Study On The Algorithm Of Texture Synthesis From Samples

Texture Synthesis is one of the research hotspots in computer graphics, computer vision and image processing, which possesses comprehensive value of applications. Texture Synthesis form Sample (TSFS) is appeared in recent years, not only TSFS can eliminate some defects of Texture Mapping technique, such as texture seams and texture distortions but also it can avoid the boring processes of parameter optimization in Procedural Texture Synthesis (PTS). Therefore TSFS is concerned by more and more researchers.TSFS develops rapidly, but there is still not a universal algorithm which can synthesize any kind of texture quickly and effectively. Better synthesis results for structural texture and the stochastic texture is gained by traditional method, but between structural texture and the stochastic texture, the synthesis results is failed by using traditional method. In the thesis, the goal is that the synthesis effect is improved for the texture between structural texture and the stochastic texture, the characteristic and the shortcomings of algorithms of TSFS are studied and analyzed, and put forward the following new viewpoints and algorithms:1 A new algorithm introducing structure information into texture synthesis is presented. The structure information of sample is recorded by integrating the edge tracking algorithm with chain code method, the structure matching error of the two patch is calculated by using the structure information of sample, the structure difference is viewed as one of the important similarity measure to ensure the structure continuity within a texture .2 A method of texture synthesis based on wavelet is presented. At first, the samples is decomposed by two-dimensional wavelet, and then the best matching patch is merely searched in the low frequency part, which enable speed up the synthesis velocity in great degree. When searching the best matching patch, the coefficients of multiresolution wavelet transform are regard as error measurement. For the sake of enhancing the matching of the texture structure information, the high frequency coefficients are endued with more weight than the low frequency coefficients when comparing the difference of the two texture patches. Whensynthesizing colored image, the existed color distortion is overcame by using KL transform.The experiment results show that not only the synthesis time is decreased in great degree, butalso the synthesis quality is improved.