一种用于图像序列拼接的角点检测算法  

Corner Detection Algorithm for Image Mosaic

作　　者：冯宇平[1,2] 戴明[1] 张威[1,2] 王美娇[3] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,长春130033 [2]中国科学院研究生院,北京100039 [3]长春理工大学光电信息学院,长春130012

出　　处：《计算机科学》2009年第12期270-271,293,共3页Computer Science

基　　金：国防科技预研基金项目(No.1040603)资助

摘　　要：分析了Harris角点检测算子的实现原理及其用于图像序列拼接时存在的不足,从几个方面提出了改进:用新的x和y方向梯度算子求取图像导数,对灰度变化更敏感;改进了角点响应函数,避免了k值选择的随机性;根据图像序列首图像自动确定R阈值,提高了算法的自动处理能力和速度;根据目标像素点的8邻域范围进行亚像素精确定位,角点位置更加准确。实验结果表明,改进算法无需人工干预,特征点提取的速度比传统算法提高了约61.3%,能更好地适应图像拼接的实际应用,是一种快速有效的角点检测方法。An improved algorithm for corner detection was proposed by analyzing the theory of Harris operator and its disadvantage. The main improved aspects were suggested just as follows： First, the new grads operators were used to get image derivative. Second,the method adopted improved corner response function for avoiding the randomicity of κ value. Then this algorithm automatically ascertained comer extraction threshold based on the first one in image sequence. Last, in order to improve the precision for corner extraction, the method obtained sub-pixel comers within 8 neighbor-hoods of the target pixel. Experiment results show that the computation time of improved algorithm enhances about 61.3% ,and it can be applied to image mosaic better.

关 键 词：图像拼接 角点检测 HARRIS算子 角点响应函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]