【作者】 许斌；

【导师】 师汉民； 唐立新；

【作者基本信息】 华中科技大学 ， 机械电子工程， 2008， 博士

【摘要】 自由曲面的三维重构技术在制造工业、艺术设计及医疗整形等领域具有广泛的应用价值。从单幅图像的灰度信息重构物体形貌(Shape from Shading,SFS)已成为计算机视觉的一个重要分支,是实现自由曲面三维重构的手段之一。但传统的SFS方法重构精度低、普适性差,而且局限于理想的漫反射模型。因此,本文将系统理论中的自组织原理引入到SFS中,将图像中的每个象素视为具有决策能力的生命单元,以灰度图像为模板,在此模板引导下,模仿生物群体的自组织行为,实现自由曲面的三维重构。本文首先对自由曲面自组织重构方法的可行性进行了研究。该方法将自由曲面视为由若干微观曲面组成的宏观曲面,并将每个微观曲面当作具有生命力的自组织单元。若微观曲面的形貌为已知,则在自组织系统的机制作用下,利用相邻自组织单元之间的短程通信以及决策能力,按照一定的演化规则,由微观曲面的形貌进行自组织迭代,逐渐重构出宏观的自由曲面。由于自组织系统具有从简单到复杂、从粗糙到细致的进化能力,因而基于自组织原理的自由曲面重构方法具有较强的抗噪性,能够消除微观曲中的一定误差。本文提出了三种理想漫反射模型下的自由曲面自组织重构方法,即基于三角片化的自由曲面自组织重构方法、基于二次曲面拟合的自由曲面自组织重构方法和基于表面法线调整的自由曲面自组织重构方法。基于三角片化的自由曲面自组织重构方法将图像中的每个象素视为自组织系统中的生命单元,每个生命单元与其相邻的两个单元组成一个三角片,该生命单元调整自己的状态,使其与邻域单元构成的三角片按朗伯反射模型(Lambertian reflectance model)满足原始灰度图像的要求。所有生命单元进行并行调整、协同处理,从而重构出自由曲面。基于二次曲面拟合的自由曲面自组织重构方法在待重构的自由曲面上某点的邻域内,依据图像的灰度模板,按照朗伯反射模型拟合一个二次曲面,并将该二次曲面视为自由曲面在该点处的微观模型。将拟合所得的每个微观曲面视为具有生命力的自组织单元,在自组织系统的内在机制作用下,重构得到宏观曲面。基于表面法线调整的自由曲面自组织重构方法改变了重构结果的表达方式,将自由曲面的表面法线视为生命单元。该方法首先对自由曲面的表面法线进行初始化,然后依据图像模板,按照自组织系统的内在演化规则,在灰度梯度约束、光滑约束、可积性约束和灰度约束的限制下,对自由曲面的表面法线进行调整,逐渐得到真实的表面法线。然后根据每个生命单元及其邻域内的表面法线拟合二次曲面,并将此二次曲面视为微观模型,再根据自组织系统,重构出宏观曲面。为了提高SFS的普适能力,本文将自组织重构技术扩展到一般光照模型下。首先建立物体表面反射特性的测定装置,用来测量物体表面的反射特性。按照Phong光照模型,根据测定的反射特性参数,对物体表面的光照成分进行分离,得到漫反射成分和镜面反射成分,再利用其中的漫反射成分来进行自由曲面的重构。实验结果表明,本文的自由曲面自组织重构方法具有较高的重构能力、较好的普适性,且可以实现一般光照模型下的自由曲面重构。该方法首次将自组织原理引入到SFS方法中,具有一定的理论价值;同时将SFS技术扩展到一般光照模型下,因而具有一定的实用价值。本文研究结果进一步完善后,可以应用于地貌航测或者月球地貌的重构、整形医疗中基于图像的三维头像或者胸像恢复、军事中移动机器人的精确导航。因此该方法能够应用于军事、医疗、艺术以及太空探测等领域,具有一定的学术价值和应用前景。更多还原

【Abstract】 The three-dimension reconstruction technique of free-form surfaces has been widely used in many fields such as manufacturing industry, art design, medical treatment and plastic surgery. Shape from shading (SFS), an important branch in computer vision field, is one of reconstruction methods of free-form surface. In order to overcome some disadvantages of traditional SFS methods such as low reconstruction precision, limited compatibility and constrained application under Lambertian reflectance model, this dissertation studies and proposes a novel SFS method based on self-organization principle in system theory. This method takes the intensity image as a target template, treats each pixel of an image as a bion that possesses decision-making capacity, and simulates the self-organization behavior of biotic populations to reconstruct the free-form surfaces guided by the target template.This dissertation uses a novel SFS method based on self-organization principle to reconstruct surface. First the feasibility of this method is studied. The free-form surface to be reconstructed is regarded as a macro-surface which is composed of some micro-surface. By using the mechanism of self-organization system, each micro-surface is treated as a bion that possesses decision-making capacity, and uses its local communication and decision-making capacity to gradually realize the self-organizing reconstruction for the free-form macro-surface guided by the transition rule. Since self-organization system has the ability of evolution from simplicity to complexity, and from roughness to fineness, and it has antinoise capacity, this method can eliminate some error caused by inaccurate micro-surface.Then based on self-organization principle, three SFS methods under idea Lambertian reflectance model are presented, namely self-organization freeform reconstruction method based on triangulation, self-organization freeform reconstruction method based on quadric surface, and self-organization freeform reconstruction method based on adjusting needle map. In self-organization freeform reconstruction method based on triangulation, each pixel of an image is taken as a bion of self-organization system, and each bion and its two neighbors constitute a triangular surface patch, then this bion adjusts its state in order that the triangle element satisfies the Lambertian reflectance model. All bion can parallelly operate and cooperate with its neighbors, and gradually reconstruct the free-form surface.In self-organization freeform reconstruction method based on quadric surface, the surface to be reconstructed can be divided into a union of micro-surface. According to the intensity target template of image, each micro-surface is approximated by a quadric surface under Lambertian reflectance map. Then each approximated micro-surface is treated as a vital self-organization unit, and all units reconstruct the macro-surface under the internal machanism of self-organization system.The reconstructed results are expressed as surface normal in self-organization freeform reconstruction method based on adjusting needle map. In self-organization system, each surface normal is treated as a bion. At first all surface normals are initialed, then guided by the image template, surface normals are parallelly and gradually adjusted by a procedure which includes three constraints: smooth constraint, intensity gradient constraint and intensity constraint. And then micro-surface is approximated by a quadric surface from each surface normal and its neighbors, in the same way, the macro-surface can be reconstructed from the known micro-surface.In order to improve the compatibility of SFS, the presented SFS technique is extended to general reflectance model. This study develops a measuring apparatus for the reflection model parameters, and then uses the parameters to separate reflection components from grayscale image, and then reconstructs free-form surface from the diffuse component.The experiment results demonstrate this SFS approach has better precision and compatibility, and can be used under general reflectance model. This method introduces self-organization principle to SFS method for the first time, and extends its application to general reflection model, so it has estimable theory value. After being further improved, the results of this dissertation can be applied to many fields, such as aerial survey of relief of Earth’s or Moon’s surface, reconstruction of 3D head skeleton or breast from image in medical aesthetics, automatic navigation of mobile military robot, so it can be applied to military, arts, medical treatments, space exploration and so on, and this method is valuable for academic research and application.更多还原