基于局部灰度梯度特征的图像快速配准方法  

Rapid image registration based on local grey scale gradient characteristics

作　　者：温江涛[1,2] 王伯雄[1] 秦垚[1] 

机构地区：[1]清华大学精密仪器与机械学系,北京100084 [2]燕山大学仪器科学与工程系,秦皇岛066004

出　　处：《清华大学学报（自然科学版）》2009年第5期673-675,共3页Journal of Tsinghua University(Science and Technology)

摘　　要：为了提高图像配准效率,对于机器视觉检测中相邻帧图像中的目标区(ROI)刚性偏移配准问题,提出了一种基于局部灰度梯度特征的快速算法。在样本和待配准图像相同几何方向上,先以方向边缘点检测算子检测目标区的数个边缘点作为特征点。以各边缘点为脊点,分别在其两侧对称求取有限数量像素的一维灰度梯度矩阵,并由此构建二维灰度梯度矩阵。根据两幅图像间的灰度梯度矩阵匹配程度来确定目标区的偏移量,进而配准图像。将本算法用于医用安瓿瓶可见异物检查图像的目标区配准过程,实现了图像目标区的无差配准。Image registration efficiency is enhanced by a rapid image registration algorithm based on the local grey scale gradient characteristics for ROI registration of neighboring images in machine vision detection. When the sample to be registered has the same view direction as neighboring images. The algorithm first detects some edge points as characteristic points with a directional edge-detection operator. The 1-D grey scale gradient matrixes of a finite number of pixels distributed symmetrically on both sides of each edge point with each edge point as a spine point are combined into a 2-D grey scale gradient matrix. The offset of the object matrix is then determined by matching the 2-D grey scale gradient matrixes to realize the image registration. The algorithm is being used for ROI registration for medical ampoules impurity inspections error-free image ROI registration.

关 键 词：图像配准 边缘点检测 灰度梯度矩阵 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]