基于活动轮廓模型的图像特征提取方法研究

【作者】 丁炜；

【导师】 张湘伟；

【作者基本信息】 广东工业大学 ， 机械设计与理论， 2004， 硕士

【摘要】 轮廓提取是计算机视觉的重要研究课题，其在虚拟现实、自控车辆、机器人环境分析、监控系统中的物体跟踪及识别、生物医学图像处理、工业在线自动检测等方面有着广泛的应用前景。在制造领域，通过对实物或模型的测量构造物体的几何模型，进而用于改进设计与制造，又称为反求工程技术。反求工程技术广泛应用于科学研究、工程技术与生物医学等领域，以解决或辅助解决几何尺寸度量、变形测量、振型测试、损伤测定、产品质量监控、实物仿形、CAD／CAM以及医学诊断等诸多问题。在用于轮廓提取的技术中，可变形活动轮廓是目前国内外研究的热点。 可变形活动轮廓模型的提出给传统的计算机视觉理论及应用研究带来了新的观点和思维方式，并已发展成为计算机视觉与模式识别中最为活跃和成功的研究领域之一。因此，基于可变形活动轮廓模型的图像处理技术研究在科学研究和工程应用中有着重要的意义。本文选题于国家自然科学基金资助项目“由一般光照条件下的图像生成实体形态的理论与方法研究”(编号：59975057)，系统的分析了国内外关于可变形模型理论与应用研究的基础上，提出了一种基于动态规划法的B样条主动轮廓模型，作为应用例之一，利用它进行人脸图像的处理和提取断层图像数据建立三维模型。其主要工作和所得到的结论如下： (1)从几何模型、尺度空间、算法实现等三个方面对可变形活动轮廓模型的理论研究与发展进行了综述；对经典的可变形活动轮廓模型的物理本质和具体求解过程进行了探讨；指出了经典可变形活动轮廓模型理论上存在的问题及进一步研究的方向。 (2)对活动轮廓模型的尺度空间进行了改进，提出了基于多尺度和多分辨率的可变形活动轮廓模型，该模型能有效的防止变形活动轮廓模型收敛于局部极小值，并且提高了变形活动轮廓模型的捕捉能力，由于低分辨率时所需控制点少所以收敛速度快。 (3)提出了一种基于动态规划法的B样条可变形活动轮廓模型。B样条方法是当前自由曲线曲面造型最为常用的方法，动态规划为解决离散最优化问题的有效手段。该模型采用控制点控制的三次B样条曲线段来表示活动轮廓，然后将控制点代入动态规划法求解，从而实现对图像的轮廓提取。该模型结合了B样条方广东工业大学工学硕士学位论文法和动态规划法的优点，可以快速稳定地收敛到目标轮廓，得到的B样条轮廓有利于进一步的表面重建处理。更多还原

【Abstract】 Contour extraction is an important problem in computer vision, and can be used extensively in many fields such as virtual reality, autonomous guided vehicles, robot environment analysis, object tracking and recognition in monitor system, biology medical image processing, industry online automatic checking and reverse engineering etc. In the field of manufacture, the geometric models are constructed through measuring them, and then they will be used to ameliorate the design and manufacture, this process is called reverse engineering. Reverse engineering is used widely in the field of science study, technology, biomedicine etc, and is used for settling the problem of geometric size measurement, transmogrification measurement, scathe measurement, surveilling the quality of manufacture, profile modeling, CAD/CAM, and physic diagnosis etc. In these techniques, the deformable active model is a hot point research.The deformable active models bring a new viewpoint to traditional computer version, and become one of the most active and successful fields in computer vision and pattern recognition. Therefore, the research of image manipulation based on the deformable active models is important signification of theory and application. In this dissertation, based on the systematically analysis of theory and application research on The deformable active models, a new B-spline deformable active models based on the DP approaches is proposed, and then it is applied to the human face detection and 3D reconstruction based on layer data. The main work and conclusion in this dissertation are as follows:The research, development and application of deformable active contour model is reviewed from geometry representation, scale-spaces and optimization method; the physical essence and the idiographic calculating process of the classical deformable active contour model are also discussed; and the existing problems and possible future research orientations arepresented.The scale-spaces of deformable active contour model has been improved, and a new model based on the MS-MR is presented, which can avoid the local minimum effectively and enforce the capture ability of the deformed active contour model. The convergence speed also has been increased because of few control points.A B-spline deformed active contour model based on DP algorithm is presented. B-spline is the most popular method to present free-form curve and surface, and DP algorithm is the effective approach for solving discrete optimization problems. In this model, the deformed active contour is represented by the cubic B-spline curve segments which are restricted by the control points, and the control points are applied to the calculation of DP algorithm, in the end the contour of image is found. Experiment results show that this model could effectively combine the merits of B-spline and DP algorithm, yielding stable, accurate and faster convergence, and its result is favorable for surface reconstruction.更多还原