项目名称: 基于目标星形先验表达及变分模型优化的自然图像场景划分
项目编号: No.61305044
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 刘李漫
作者单位: 中南民族大学
项目金额: 25万元
中文摘要: 自然图像场景的区域分割是许多计算机视觉研究问题的基础及难点,当前大多数图像分割方法由于缺乏先验知识容易导致分割结果盲目,而基于学习的方法又存在样本选择困难等问题,因此,本课题提出研究集成目标星形先验表达及变分模型优化的自然图像场景划分方法。主要解决三个方面的问题:星形形状先验约束变分模型的构建;目标星形先验中心的自适应获取;多变分模型的快速准确优化。首先结合分段曲面内部平坦性、分段曲面边界光滑性、曲面拟合前后相似性以及目标星形形状先验等多种约束,考虑能量函数的子模特性及图切分测度理论,构建多类场景自动划分变分能量模型。进而研究有效的目标星形中心自适应获取方法,构建目标星形形状先验表达数学模型。最后研究有效的区域自适应重标记方法,探索等价图的分解方法及图切分合并算法,研究高效的多层权值图能量函数映射表达,解决多变分模型的高效准确优化问题。研究成果不仅具有重要理论意义,其应用前景也十分广阔。
中文关键词: 图像分割;星形先验;变分模型;快速优化;
英文摘要: The partition of natural image scene is the fundation and difficulty of most of the computer vision research. To solve the blindness derived from the lack of the prior knowledge in image segmentation and the sample selection and computational complexity problems in learning-based methods, this project proposes to study the natural scene partition method by integrating the object star shape prior and variational model optimization. We focus on three aspects: 1)the construction of the variational model with the star shape prior constraint, 2) the adaptive acquisition of the star center of objects, 3)the fast optimization algorithm of multiple variational model. Firstly, by fusing the flat constraint in the piecewise surface, the smoothing constraint of the boundary of the piecewise surface, the similarity between the original surface and the fitting surface and the object star shape prior constraint, the multi-class partition variational energy function is constructed based on the submodular feature of the energy function and the cut metric theory. Secondly, the effective adaptive star center acquisition method is studied and the star shape constraint mathmatical model is constructed. Finally, we study the adaptive region relabeling method, the equivalent graph decomposition and cut composition method, the effecti
英文关键词: Image segmentation;Star Shape Prior;Variational Model;Fast Optimization;