基于角点检测的图像分割算法及在三维重建中的应用  

Corner Detection Based Image Segmentation Algorithm and Its Application on Three-dimension Reconstruction

作　　者：陈德明[1] 孙光民[1] 许磊[1] 王纯[1] 

机构地区：[1]北京工业大学电子工程系,北京100022

出　　处：《计算机测量与控制》2009年第1期173-176,共4页Computer Measurement &Control

基　　金：北京市教委科技发展计划项目(KM200710005009)

摘　　要：针对基于单幅图像的三维重建方法的多解性和病态性的难点问题,提出了一种基于Harris多尺度角点检测的图像分割算法,将复杂的工程图像分离成若干个简单基本几何形体,分别对其重建以避免直接恢复深度信息的病态解问题;为了提高基于角点的图像配准算法的配准精度,把多分辨分析的思想引入到经典的Harris角点检测中,构造了基于小波变换的灰度强度变化公式,并得到了具有尺度变换特性的自相关矩阵,从而使改进的Harris角点检测算法具有旋转、平移和尺度的不变性;实验验证了改进算法的快速、准确和稳定的特性。Aiming at the multiple solutions and ill--posed problem of the single--image--based 3D reconstruction method, an image seg- mentation algorithm based on Harris multi--scale corner detection is proposed in this paper. At first, the complicated engineering drawing can be split into several simple geometrical shapes. Then the system will reconstruct the 3D information for each simple geometrical shape and the ill-- posed problem can be avoided as a result. The performance of feature detector determines the precision of registration directly. In this paper, the multi--resolution idea is introduced into the classical Harris algorithm, and the wavelet-- based formula for measuring the image intensity variation is developed, meanwhile, the auto--correlation matrix is obtained that reflected the scale variation information. So, the features of the improved Harris multi--scale corner detection algorithm are invariant to translation, rotation and scale transformations. Ex- perimental results show that the proposed algorithm is fast, accurate, and robust.

关 键 词：三维重建 角点检测 图像分割 图像匹配 

分 类 号：TN911.73[电子电信—通信与信息系统]