A Pipeline Architecture For Real-Time Connected Components Labeling

Abstract

(1966)) and architecture were verified by software simulation. The transitive closure label equivalence process isperformed by a content addressable memory. The scheme takes full advantage of the concurrent memory operations providedby content addressable memory, and performs the connected components labeling in only two pipeline frames, independent ofthe complexity of component shapes in the input image.The connected components labeling module can work in conjunction with the existing feature and moment extractionhardware. The pipeline based architecture allows other image processing operations to be performed in the same pipelinepreceding the connected components labeling module. Thus, the connected components operations effectively take no additionaloperation time. This simple architecture should be low cost and easy to implement in hardware.I. INTRODUCTIONThe operations of connected components (CC) labeling and feature extraction convert image data into symbolic featurelists (see Rosenfeld et al., 1966, 1976 and Lumia et al., 1983) and so play a crucial interface role in any intelligent visionsystem. The CC labeling operation assigns each connected region of a binary image a unique label. The CC feature extractionoperation extracts such region features as centroid, area, moments, mean, etc. This labeling and characterization forms thebasis for the high -level intelligent vision processing. Regions and their feature data can be stored as schemata in a high -leveldatabase. Which is used instead of pixel data by high -level processing unit. Real -time CC processing is a prerequisite for anycomputer vision system to achieve real -time performance.In many existing machine vision systems, the CC labeling and feature extraction tasks are performed by software packagesrunning under a general purpose CPU (see Shapiro et al. 1985 and Gleason et al., 1979) and CC analysis is limited by slowoperations of the sequential host computer to images containing a small number of components or to non -real time applications.Recently, a connected components feature extraction hardware board set is marketed by Vision system Inc.(see APA512 -MXuser's manual, 1987). Unfortunately, the hardware can only provide connected component feature lists rather than a connectedcomponent labeled image.Shapiro, Lee, and Haralick (1985) proposed a real -time CC labeling module based on a state machine architecture. The