摘要
SUSAN角点检测算子的提出是以假设待测角点是L型为前提的,这就造成了SUSAN算子在检测角点时以USAN区域的大小为判据的局限性。实际上,当USAN区域的大小等于SUSAN圆模板面积的一半的时候,常常会出现错误的检测结果。在分析图像各特征点的本质区分的基础上,在SU-SAN圆模板内,附加了一个圆环模板,并以圆环模板上灰度的跳变次数为辅助判据,来弥补SUSAN算子的不足。此外,SUSAN算子USAN区域的划分是基于固定灰度差阈值的,这对于具有不同对比度的图像的角点提取很不利。鉴于此,提出了一种基于迭代运算的灰度差阈值的计算方法,在每个像素位置,通过迭代运算计算其对应SUSAN圆模板内的灰度差阈值,得到更合理的USAN区域。所提出的算法以USAN区域大小为第一判据,再辅以圆环模板上的灰度跳变次数为第二判据,从而为特征点的检测提供了双重保障。实验结果表明,算法可以准确、可靠地提取出各种不同类型的角点。
The SUSAN（Smallest Univalue Segment Assimilating Nucleus） corner operator is proposed under the assumption that the corners to be detected are L-shaped,which results in SUSAN operator′s limitations in using USAN（Univalue Segment Assimilating Nucleus） region′s size as the criterion.In fact,wrong detections often happen when the USAN region′s size is equal to half of the area of the SUSAN circular mask.A ring-shaped mask was attached within the SUSAN circular mask based on the analysis of essential distinction of various image features and the times of intensity change was used as the criterion to overcome the deficiency of SUSAN operator.In addition,the USAN region is obtained by using a fixed brightness difference threshold,which is disadvantageous for corner detection with different contrast image.Therefore,we propose an iterative calculating method for brightness difference threshold,and calculates the brightness difference threshold of the corresponding SUSAN circular mask at each pixel location by iterative operation whereby to obtain a more pretty USAN region.The proposed algorithm provides double assurance by using the size of USAN region as the first criterion and supplementing the times of brightness change as the second criterion.Experimental results show that the algorithm can accurately and reliably extract various types of corners.
出处
《机械科学与技术》
CSCD
北大核心
2011年第7期1120-1123,共4页
Mechanical Science and Technology for Aerospace Engineering
基金
国家重大专项项目(2009ZX04001-065)资助
