项目名称: 基于凸优化理论的特征点匹配算法研究
项目编号: No.61301269
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 无线电电子学、电信技术
项目作者: 李鸿升
作者单位: 电子科技大学
项目金额: 24万元
中文摘要: 特征点匹配是计算机视觉中一个重要的问题,其在图像拼接、图像检索、目标检测与识别等方面具有重要的研究意义。现有匹配算法一般仅支持特定的几个几何变换模型,限制了算法的应用范围。本项目研究一种基于凸优化理论的匹配算法框架,其支持所有可被表达为仿射函数的几何变换模型,能够扩展匹配算法的适用范围和易用性。针对凸优化理论下匹配算法的一些限制和难点,提出有效的解决方案:(1)为特征点匹配问题建立恰当的凸优化数学模型,研究相应的相似度度量函数。(2)为计算机视觉中的常见应用,设计可被仿射函数表达的几何变换模型,在该算法框架下高效的解决这些问题。(3)针对建立的凸优化模型,研究一种高效的优化策略求解其全局或近似全局最优解。
中文关键词: 图像特征匹配;物体匹配;形状匹配;凸优化;凸松弛
英文摘要: Feature matching is an important problem in computer vision. It has extensive uses in image stitching, image retrieval, and object detection and recognition. Existing matching methods only support several specific geometric transformation models, which limit their uses in different applications. This project researches a feature matching framework based convex optimization techniques. It supports all geometric transformation models that can be expressed by affine functions. This property would improve the usability of feature matching methods. To mitigate limitations of the convex-optimization-based feature matching methods, this project proposes several solutions: (1) the proposed method defines appropriate similarity metric functions to design a proper convex optimization model for feature matching problems. (2) For different computer vision applications, several geometric transformation models that can be expressed by affine functions are designed. (3) An optimization strategy is proposed to efficiently recover the model's global or approximately global optimum.
英文关键词: feature matching;object matching;shape matching;convex optimization;convex relaxation