摘要
目标特征是目标分类识别的基础,在二值图像中常基于像素连通关系进行提取并通过连通域标记使所含像素能便捷访问。文章简要分析了常见连通域标记算法的性能,针对三种特殊而常见的连通域与目标间的对应关系指出其在目标特征提取中的不适用性,为提高目标特征完整性和准确性在目标语义层次重新定义了像素连通,利用区域边界扫描以及对边界像素的有效标记,将取决于二维信息的连通关系识别转化到一维中,使每行(列)的处理过程相互独立,从而实现一遍逐行(列)的图像扫描即可完成特征提取(包括目标轮廓)及连通域标记,且标号连续。最后通过仿真从效率、适用性及内存耗费等方面对算法进行了验证与分析。
The target feature is the foundation of target classification and recognition, which can be extracted based on pixel connected relation and accessed easily by pixel through connected component labeling in binary image. This paper briefly analyzes the function of common connected component labeling algorithm, points out the inapplicability of connected component labeling algorithm in target feature extraction aiming at three kinds of special and common corresponding relations between connected components and targets, and redefines the pixel connectivity in the target semantic level to improve the completeness and accuracy of target feature. Using the regional boundary scanning and the efficient labeling, the connected relation depending on two-dimensional signal is recognized and transformed into one-dimensional to make the processing procedures in each row (column) mutually independent, so that the feature extraction (including target contour) and the connected component labeling are realized through picture scanning row by row (line by line), the labeling number is continuous. Finally, through simulation, this paper verifies and analyzes the algorithm from such aspects as efficiency, applicability and memory consumption.
出处
《计算机与网络》
2015年第7期58-61,共4页
Computer & Network
关键词
连通域标记
轮廓跟踪
特征提取
connected component labeling
contour tracking
feature extraction