Texture Synthesis Algorithm Based On Classified Blocks Quilting

Textures can describe a wide variety of natural phenomena with random variations over repeating patterns. Since reproducing the realism of the physical world is a major goal for computer graphics, textures are important for rendering synthetic images. However, it is difficult to describe and reproduce textures under a common framework because they are so diverse. Texture Synthesis is a method to solve this problem. Synthesis has been an important research topic in Computer Graphics and Computer Vision in recent years. It can be useful in a lot of applications, such as data compression, textures transmitting over network, Computer Animation, restoration and editing images, etc.The traditional ways of Texture Synthesis from Samples (TSfS) analyses the texture using the statistical tools, but they just can process the random texture in most cases; the Pixel-Based TSfS algorithms can improve the quality of result, but extremely slow; the Patch-Based TSfS algorithms, which use the relativity of neighborhood, overcome the limitation of efficiency, and get the satisfied results. These methods also expand the range of synthesized textures. So the Patch-Based Synthesis is the direction of development for TSfS.We study the methods of Patch-Based Texture Synthesis and find that these kinds of methods have their limitations: they can't keep the integrality and continuity of the original textures, they have weak randomicity and they can repeat patches incidentally. The repetition is more distinct when the size of sample texture is small and the result is big. Aim at these problems, we present the synthesize algorithm based on the classified blocks. Unlike previous texture synthesis algorithm, blocks of input texture are first classified. The structure of the classified blocks of the input texture is formed and patched to the output texture as a basic unit. Our method synthesizes a new image by stitching together the basic structures of each class, but not square of regular size, like previous algorithms do. Notwithstanding its simplicity, this method is a special-suited for existing textures.Then we introduce the idea of classified blocks to the synthesis of video textures. We map the video to a cube by the axis of time. Then blocks of input texture are first classified. The structure of the classified blocks of the input texture is formed and patched to the output texture as a basic unit. This measure is similar to the 2D image and the only difference is the block here is not tabulate any more.