Research On The Key Technologies Of Visual Media Editing And Reusing

In recent years, with fast developing of the internet, there are abundant resourcesof image/video. At present, these vast amounts of images/videos have been widelyused in the fields of virtual reality, film making, computer games, animation creation.It was widely reused in film, television and animation production by editing and pro-cessing the existing image/video materials, which not only improves the level andutilization of image/video, but also greatly reduces the costs. The efficient reusingof image/video is based on the research for constructing the image/video’s editingtoolsets. Accordingly, the deeply developing, rapidly processing, effectively usingof image/video materials thus are meaningful. On the one hand, it can increase theamount of image/video resources for film, television and animation production. Onthe other hand, it improves the efficiency of film, television and animation productionby the reusing the image/video resources.This dissertation focuses on some issues of image/video editing and reusing, andwe study some key issues about2D texture synthesis, fluid animation simulation, andcolorization by analyzing the inherent structure of the texture and using the texturefeature of image/video. And these related algorithms were implemented. The maincontributions are summarized as follows:(1) A fast texture synthesis method based on feature description is proposed.Patch-based texture synthesis techniques from a small sample texture generate a newtexture image efficiently, but one significant problem is that these existing texture syn-thesis algorithms suffer from structure mismatching across patch boundaries due toinaccurate similarity measure. We propose a desired distance metric to measure struc-tural similarity when especially handling textures with salient features, which has theability to find best matches during the patch-based sampling. To do this, we introduce the concept of”layers” to describe the correlation of pixels through computing eachpixel weight, and construct the orientation fields to capture the inherent direction andvariation of texture features through computing the dominant local orientation angleand the degree of anisotropy of each pixel. Our experimental results demonstrate thatthe proposed method can significantly alleviate structure discontinuities in synthesizedtextures.(2) A texture synthesis and design method is proposed. Since human perception issensitive to structure and periodicity, we thus perform texture synthesis by taking intoaccount the distribution of texels. Given a sample texture, the analysis procedure con-sists in segmenting texture into individual texels, and detecting each texel in order toanalyze their neighborhood relationships by constructing connectivity. Then the syn-thesis process consists in reproducing a new large texture directly on a user-specifiedcanvas by recomposing segmented texels, which synthesizes two-dimensional texel ar-rangements based on the previously constructed neighborhood relationships of texels.Results show that the method especially deals with the near-regular textures, whichwell preserves underlying structural regularity.(3) A method for fluid animation from a single image by editing video examplesis proposed. We perform synthesizing animation from a single image by transfer-ring fluid motion of a video example globally. Given a target image of a fluid scene,an alpha matte is required to extract the fluid region. Our method need to adjust auser-specified video example for producing the fluid motion suitable for the extractedfluid region. Employing the fluid video database, the flow field of the target imageis obtained by warping the optical flow of a video frame that has a visually similarscene with the target image according to their scene correspondences, which assignsfluid orientation and speed automatically. The results demonstrate that our method issuccessful in preserving large fluid features in the synthesized animations.(4) A method for grayscale image colorization by using parallel image segmen-tation and the related applications is proposed. Given a grayscale image to colorize,we construct the relationships between pixels in the input images and the regions ex-tracted from a segmented example image by using a robust supervised classificationscheme, assigning color values for these pixels. Combined with spatial filtering with the spatial coherence, the colors of pixels with highest confidence values are retained.This is provided as color palette to the grayscale image. In addition, the regions withsimilar features in input image can be segmented in parallel by using texture basedvariational level set method. It can fast and accurately determine the boundaries of theregions. With the segmented regions, various colorization techniques can be appliedto replace colors. We also apply this parallel image segmentation for repeated sceneelements, and edit images by using replacement and rearrangement operations.