A two-step algorithm and its parallelization for the generation of minimum containing rectangles for document image segmentation

In document processing, segmentation is done to uniquely identify each foreground connected region of an image by specifying its minimum containing rectangle (MCR). MCR is the rectangle with minimum dimensions that completely encloses a geometric pattern. We present a two-step MCR detection algorithm and its parallelization method. The first step determines the boundary of each connected component in a document image. This reduces resource requirements and speeds up the subsequent rectangle detection step. The rectangle detection step determines MCRs of the connected components from the detected boundaries. A comparison is made between a single-step and the two-step approaches of MCR detection. Both the boundary detection and the rectangle detection steps are parallelized and implemented on transputers to reduce the total processing time.