摘要
在工业、军事等领域,有时需要检测图像感兴趣区域是否有异常发生。而抖动和光照变化都会导致异常的误判,消除它们实现精准的异常检测是一个难题,为此,提出一种解决办法。该方法使用基于投影的快速分块匹配来消除待检测图像与模板图像间可能存在的抖动;接着多项式拟合有问题部分的光照变化来消除其影响;最后基于加权多区域差分矩阵进行异常的判定。研究结果表明：此方法能精准地检测到图像感兴趣区域内异常的发生。
In industry and military field,sometimes it is necessary to detect whether there is abnormity in the interesting areas of the image.It is a difficult task to eliminate vibration and illumination variation which could cause wrong abnormity judge to realize precise abnormity detection,so a new method is presented.This method uses fast block-matching based on projection to eliminate possibly existing vibration between the detecting image and the template image;And then polynomial fitting illumination variation of problem places to eliminate its influence;Finally carrying out abnormity decision based on the weighted multi-areas difference matrix.The study result shows that this method can precisely detect the abnormal changes of the interesting areas in the image.
作者
张果胜
ZHANG Guo-sheng（School of Information Science and Engineering,Central South University,Changsha 410083,China）
出处
《电脑知识与技术》
2010年第5期3489-3491,3494,共4页
Computer Knowledge and Technology
关键词
异常检测
块匹配
多项式拟合
加权多区域差分矩阵
abnormity detection
block-matching
polynomial fitting
weighted multi-areas difference matrix
