基于运动迁移的图像非刚性匹配与特征点提取方法研究

项目名称： 基于运动迁移的图像非刚性匹配与特征点提取方法研究

项目编号： No.61203254

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 自动化学科

项目作者： 杨旸

作者单位： 西安交通大学

项目金额： 25万元

中文摘要： 图像非刚性匹配是计算机视觉、模式识别和图像处理领域的关键技术。图像非刚性变换的复杂性与未知性，是匹配方法研究的难点。本项目借鉴图形学领域的运动迁移思想，探索获取未知目标个体,形变先验知识的有效计算途径，以此突破图像非刚性匹配算法中的关键技术。针对非理想条件下的图像匹配问题，结合基于样本学习的运动迁移模型，提出一种新的鲁棒图像非刚性匹配方法；利用局部特征匹配、空间运动约束，结合结构化的特征表示形式，研究匹配参数初始化与快速搜索方法。另外，将探讨运动单元迁移的组织建模方式，学习样本序列的运动轨迹先验知识，将图像非刚性匹配应用于人脸运动序列的特征点提取工作中。本项目的实施有助于完善基于形变先验知识的图像非刚性匹配算法理论框架，同时可为实际中海量图像数据的特征点标定与整理工作提供可靠的技术手段。

中文关键词： 图像匹配；非刚性；运动迁移；特征点提取；

英文摘要： Image non-rigid registration is a key technique in the fields of computer vision, pattern recognition and image processing. Because of the complexity and unknowns about the non-rigid image transformation, it is hard to improve the registration accuracy. This project uses the idea in the motion transfer in the computer graphics, to acquire some prior knowledge about the deformation of unknown targets, which can overcome the limitations in traditional approaches. To solve the fitting problem under unconstrained conditions, some example-based motion transfer models are integrated into image registration algorithms. We propose a novel and robust registration approach with a single template. To utilize local feature matchings, space motion constraints, and hierarchical feature structures, we present our parameter initialization and rapid registration search methods. In addition, by learning the prior knowledge about the motion tracks from sample data, a transfer model is built for motion units. In that case, the image registration will be applied for the feature extraction for complex face motion sequences. The project is a supportive research in the theory framework of deformation-prior based non-rigid image registration. In practice, it will play an important role in the applications of feature points annotation an

英文关键词： image registration；non-rigid；motion transfer；feature points extraction；