Fast connected component labeling in binary images

This paper presents algorithm for fast connected component labeling of the binary images. Simplicity of the algorithm predetermines its use especially in real-time applications like human computer interfaces. Current algorithms are based on different hierarchical structures and union-trees. The presented algorithm does not connect every component as in a common approach. Instead, some regions stay left unconnected. This property may be beneficial in some cases, especially when two neighbouring objects, in the binary image, look like one connected region. Two way testing north-left was used to determine component membership. If two different regions are detected in the neighborhood, the algorithm decides to merge or split the component regions.