一种适合硬件实现的多值图像连通域标记算法  

A connected components labeling algorithm for multi-value image that suitable for realization in VLSI

作　　者：桑红石[1] 傅勇[2] 张天序[1] 刘云生[2] 

机构地区：[1]华中科技大学图像识别与人工智能研究所 [2]华中科技大学计算机科学与技术学院,湖北武汉430074

出　　处：《华中科技大学学报（自然科学版）》2005年第9期5-8,共4页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金重大项目资助(60135020).

摘　　要：提出了一种合适硬件实现的多值图像连通域标记算法.算法以有效的方式识别、保存并整理区域之间复杂的连通关系.首先,通过对图像执行一遍逐像素扫描,得到图像的初步标记结果以及初步标记之间的连通关系;然后,通过等价表整理以及图像代换,输出图像标记的最终结果,并且连通域按照扫描顺序被赋予唯一的连续自然数.仿真结果表明,该算法能够识别图像中复杂的连通关系,产生正确的标记结果,在用于大幅图像的分块标记时,能够实现全局标记一致.在以硬件方式实现该算法时,在50 MHz工作时钟下,处理320×256像素图像能够达到100帧/s的标记速度,满足大部分实时目标识别系统的要求.A connected components labeling algorithm used for multi-value segmented image labeling that is easy to implement for special purpose VLSI chips was proposed in this paper. This algorithm can detect, record, and organize the complex connections of connected components effectively. Firstly, the image is scanned alphabetically, and then a temporary labeling image is produced. Meanwhile, connections of connected components are recorded in the connection-table. By connection-table organizing and temporary labeling image substituting, a labeled image is produced, and the connected components are labeled in natural number. Simulation results showed that the algorithm proposed in this paper can label image with complex connections correctly, and label large image correctly by partition the large image into small pieces. When this algorithm is realized in VLSI, it can label 320 × 256 images at speed of 100 frames per second. The VLSI chip can be used in many real-time pattern recognition systems.

关 键 词：多值图像 连通域标记 模式识别 超大规模集成电路 

分 类 号：TN911.73[电子电信—通信与信息系统]