摘要
本文提出一种快速连通域分析算法,它对像素的行程进行操作,并将标号作为行程及连通域的特征之一,特征通过数据结构的指针与行程及连通域相联系。该算法运用了两个关键技术,一是设计了一种链式机制来表示和实现标号的等价关系,二是通过指针的传递来实现标号及其它特征的向下传递和逆向传播,特征在标号过程中动态修改。这样甚至能实现仅对图像一遍扫描便能完成连通域标记和常用特征量的计算。实验表明了本文算法的有效性。
With current PC configuration, the memory requirements for connected components labeling algorithms are easily satisfied while the time costs usually still cannot meet practical demand. Here a fast algorithm calculating shape/grayscale features of connected components while labeling them is proposed. It is executed on pixel runs and labels are considered as one of the features, which associate to runs or objects through a data pointer. A chain mechanism is designed to represent and realize the equivalence of labels. By passing the pointers and dynamically modifying the feature data, equivalence and back propagation of labels are realized while the label and other features are passed on. The later two techniques make connected components anslysis can even be implemented through only one image scanning. Experimental results show the effectiveness of the algorithm.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2003年第1期110-115,共6页
Pattern Recognition and Artificial Intelligence
关键词
快速连通域分析算法
计算机视觉
连通域标记
链式等价机械
指针
Connected Component Analysis (CCA), Connected Component Labeling (CCL), Run Length, Chain Mechanism for Equivalence, Pointer
