基于交互分析的图像与视频快速编辑扩散技术研究

项目名称： 基于交互分析的图像与视频快速编辑扩散技术研究

项目编号： No.61202294

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 王栋

作者单位： 华南农业大学

项目金额： 25万元

中文摘要： 可视媒体编辑是提升媒体质量、实现呈现目标的重要手段，而编辑扩散是交互式可视媒体编辑的一个核心问题。目前主流的编辑技术都是用最优化方法来对扩散问题进行建模，计算效率低、占用内存空间大，不适用于处理视频这样大数据量的媒体；此外，交互方式设定比较随意，缺乏与编辑意图的关联.为达到提高编辑扩散效率同时不降低编辑的质量的目的，本课题将以图像和视频为对象研究三个方面的内容：(1) 系统地研究用户交互方式与编辑意图间的关联，提出一些交互设计准则；具体地，将考察只勾画出待改变的前景区域的像素样本，而自动计算不变区域的编辑策略；(2)拟定义一种有界双调和函数的相似性度量准则，使得扩散过程中每种编辑模式除了满足插值、光顺等基本要求外，并只作用于一个或多个相似的局部区域，且能保证不同编辑模式之间的边缘区域自然光滑过渡；(3)提出一些基于高维空间插值技术的快速编辑扩散方法，最好有闭形式的解以便并行加速计算.

中文关键词： 图像编辑；编辑扩散；细节保持；边缘感知；超分辨率

英文摘要： Visual media editing such as image editing and video editing is a key means to enhance the media quality and realize presentation goals while edit propagation is its kernel component. Most of the mainstream editing techniques formulate the problem using optimization methods. They are usually ussuitable to process media such as high videos and resolution images due to the drawbacks such as low computational efficiency and high space occupancy.Furthermore, the ineractive styles are almost casually specified in these methods and have little relationship with editing intents.To achieve the goal of improving the effiency of edit propagation and maintaining the quality results at the same time, this proposal launches an investigation on the following topics: (1) Define rules for designing interactive styles by systematically analyzing the relationship between interactive styles and editing intents; Concretely, we should study a pattern which only interactively specifies regions to be changed but automatically computes regions remaining unchanged; (2) Introduce a similarity criterion based on bounded biharmonic functions so that the edit propagation can not only satisfy basic properties such as interpolation and smoothness but also impact on multiple similar regions while maintaining natural transition across boundary

英文关键词： image editing；edit propagation；detail-preserving；edge-aware；super-resolution