Research On The Example-based Texture Design, Completion And Re-texturing

Textural phenomena are ubiquitous in our environment. Examples include tex-tures such as sound, image, video, motion, surface geometry and so on. A com-mon characteristic among textural phenomena is that their appearance is sta-tistically self-similar over space or time. Since reproducing the realism of thephysical world is a major goal for computer graphics, textures are important forrendering synthetic images and animations. For the wide applications in digitalentertainment, game design and virtual reality simulation, etc., the research ontexture has been one of the hottest topics in computer graphic, computer visionand digital image processing. However, because textures are so diverse it is dif-ficult to describe and reproduce them under a common framework. In this dis-sertation, we explore some key problems in texture technique, such as example-based image completion in gradient domain, completion-based texture designusing deformation, deformation-based interactive texture design using energyoptimization, and high dynamic range (HDR) image texture replacement, andsome new approaches are proposed:(1) Image completion. This dissertation presents a novel gradient-basedimage completion algorithm for removing significant objects from natural pho-tographs. Our method reconstructs the region of removal in two phases. Firstly,the gradient maps in the removed area are completed through a patch-based fill-ing algorithm After that, the image is reconstructed from the gradient maps bysolving a Poisson equation. A new patch-matching criterion is developed in ourapproach to govern the completion of gradient maps. Both the gradient and thecolor information are incorporated in this new criterion, so a better image com-pletion result is obtained. Several examples and comparisons are given at theend of the paper to demonstrate the performance of our gradient-based imagecompletion approach.(2) Texture design. This dissertation presents a novel approach for design-ing a variety of large textures from a single small sample texture. Firstly, theoriginal small texture is segmented into layers, each of which contains one par-ticular texture element. Secondly, each layer is deformed using a set of chaotic- based transformation operations. Thirdly, all the deformed layers are added to-gether to form a new texture, which is a natural variation of the original sampletexture. Since each layer is deformed independently, adding the deformed lay-ers together usually results in a texture with overlapping regions and holes. Weemploy the graphcut algorithm and an example-based image inpainting tech-nique to seamlessly patch the overlapping regions and to fill the holes. More-over, an optimized graphcut synthesis algorithm and a new cyclic texture syn-thesis technique are also developed for efficiently creating large seamless tex-tures. As a result, our approach shows particular strength in generating a largevariety of textures from a single sample texture while avoiding highly repetitivepatterns. Our experiments demonstrate that the proposed technique can alsobe used for other texture synthesis applications, such as texture synthesis frommultiple samples.(3) Interactive texture design. This dissertation presents a novel interactivetexture design scheme based on deformation and energy optimization. Givena small sample texture, the design process starts with applying a set of defor-mation operations to the sample texture to obtain a set of deformed textures.Then local changes to those deforined textures are further made by replacingtheir local regions with the texture elements interactively selected from othertextures. Such a deform-select replace process is iterated for many times un-til the desired deformed textures are obtained. Finally those deformed texturesare composed to form a large texture with the graph-cut optimization. By com-bining the graph-cut algorithm with an energy optimization process, interactiveselections of local texture elements are done simply through indicating the po-sitions of texture elements very roughly with a brush tool. Our experimentalresults demonstrate that the proposed technique can be used for designing alarge variety of versatile textures from a single small sample texture, increasingor decreasing the density of texture elements, as well as for synthesizing texturesfrom multiple sources.(4) Texture replacement. Based on fast trilateral filtering, This dissertationpresents a novel tone mapping and re-texturing method for high dynamic range(HDR) images. Our new trilateral filteringbased tone mapping is about 7 to10 times faster than the existing one in. Firstly, a novel tone mapping al- gorithm for HDR images is presented based on the fast bilateral filtering andtwo newly developed filters: the quasi-Cauchy function kernel filter and thefourth degree Taylor polynomial kernel filter. Secondly, a new gradient basedimage re texturing method is introduced, which consists of three steps: 1) con-verting HDR images into low dynamic range (LDR) images using our fast trilat-eral filtering-based tone mapping method; 2) recovering the gradient luminancemaps for the region to be re-textured: 3) reconstructing the final re-textured im-age by solving the Poisson equation. The proposed approach is suitable for HDRimage tone mapping and re-texturing, and the experimental results have shownthe satisfactory performance of our method.