基于距离权重积分的凸度衡量方法

发布时间：2018-01-28 14:07

本文关键词： 形状分析特征提取凸度衡量形状分类三维模型检索　出处：《华东师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：形状分析是计算机视觉领域里的一个热门研究方向,并在模式识别、图形标注、形状分解、图像配准等领域得到广泛应用。而形状提取的过程中容易出现扭曲、遮挡、噪声等干扰,所以提出一种既满足平移、缩放、旋转不变性又对噪声有较高鲁棒性的形状描述符具有重要的理论意义和实际意义。目前紧密度、线性度、矩形度、凸度等几何特征得到了国内外的广泛研究,其中凸度表示物体的凹凸程度,是一种全局几何特征,因其具有显著的视觉特性,在视觉感知中扮演着重要的角色。现有的凸度测量方法可以分为三类:基于面积的方法、基于边界的方法和基于概率的方法。其中,基于面积的方法以其计算简便、抗噪性强等优点被广泛的使用。然而现有的基于面积的方法由于只考虑凹陷面积和凸包面积的大小,导致其测得的许多结果并不合理。本文提出了一种基于面积的二维凸度衡量方法,它是对原始基于面积的方法的一种改进。本文假设所有非凸的形状都是由其凸包不断凹陷所致,不同的凹陷方式会对原形状产生不同的影响力,如果凹陷对原形状有较大的影响力,则形状具有较低的凸度值,反之,则形状具有较大的凸度值。我们采用了一种距离权重积分(Distance Weighted Area Integral,简称DWAI)的方法来计算凹陷的影响力,根据每个点与形状的凸包中心(Geometric Center of ConvexHull,简称GCCH)的距离来分配这个点的影响力,离GCCH较远的点具有较低的影响力,即在形状的凸包边缘的凹陷区域具有较低的影响力。通过改变参数影响因子可以调整区域的影响力,增大影响因子可以增大凹陷位置带来的影响,反之则减小影响,当影响因子等于0的时候,本文的方法退化为原始的基于面积的方法,即凹陷的位置不影响凸度的计算,所以本文的方法可以完全取代原始的基于面积的方法。其次,通过把DWAI的思想拓展到三维空间,本文提出了一种三维凸度衡量方法,视觉感知上比现有最新的Lian的方法更加合理。根据此方法,通过设置不同的参数,提取统计信息,本文设计了一种基于凸度的三维模型形状描述符CS(Convexity Statistic),与Lian的基于凸度的形状描述符CD(Convexity Distribution)相比,CS总体上具有更好的检索效果。本文通过理论证明验证了所提方法的有效性,大量的实验结果表明,本文的方法在定性和定量两方面都优于其他对比方法。
[Abstract]:Shape analysis is a hot research field in the field of computer vision, and has been widely used in the fields of pattern recognition, graphics annotation, shape decomposition, image registration and so on. Therefore, a shape descriptor which not only satisfies translation, scaling, rotation invariance but also has high robustness to noise has important theoretical and practical significance. At present, the degree of compactness and linearity is very important. Rectangle, convexity and other geometric features have been widely studied at home and abroad, in which convexity represents the concave and convex degree of objects, which is a global geometric feature, because of its obvious visual characteristics. It plays an important role in visual perception. The existing convex measurement methods can be divided into three categories: area-based method, boundary-based method and probability-based method. The area-based method is widely used because of its simple calculation and strong anti-noise. However, the existing area-based methods only consider the size of the concave area and convex hull area. As a result, many of the results obtained are unreasonable. In this paper, an area based two-dimensional convexity measurement method is proposed. It is an improvement on the original area-based method. In this paper, it is assumed that all non-convex shapes are caused by continuous indentation of its convex hull, and different indentation modes have different effects on the original shape. If the depression has a greater influence on the original shape, the shape has a lower convexity value and vice versa. We adopt a distance Weighted Area Integral. The method of DWAI is used to calculate the influence of the depression, according to the center of the convex hull of each point and shape of the geometric Center of ConvexHull. The distance between GCCHs to distribute the influence of this point, the point farther away from the GCCH has lower influence. The influence of the region can be adjusted by changing the parameter influence factor, and the influence of the depression location can be increased by increasing the influence factor. When the influence factor is equal to 0, the method in this paper degenerates into the original area-based method, that is, the position of the depression does not affect the calculation of the convexity. Therefore, this method can completely replace the original area-based method. Secondly, by extending the idea of DWAI to three-dimensional space, this paper proposes a three-dimensional convexity measurement method. Visual perception is more reasonable than the latest Lian method. According to this method, the statistical information is extracted by setting different parameters. In this paper, a 3D model shape descriptor based on convexity, CS(Convexity Statistics, is designed. Compared with Lian's Convexy-based shape descriptor CD(Convexity Distribution. In this paper, the effectiveness of the proposed method is proved by theory. A large number of experimental results show that the proposed method is superior to other comparative methods in both qualitative and quantitative aspects.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】