Color document image segmentation for automated document entry systems

Monochrome document image segmentation has been studied for over ten years. On the other hand, how to segment color document images is still an open research field. We propose an approach for segmenting color document images. Unlike the common practice in monochrome documents that objects are black on a white background the components in color documents can be any color. To cope with their variety, the first step of our approach is to create a binary image of edge-representation. Then page segmentation is carried out in the binary image using the CRLA procedure. Finally, we utilize the geometric features and the color information to classify the segmented blocks into text lines and picture components. The identified text lines are then further transformed into the white-background/black-text format for OCR processing. The proposed approach was implemented on a Pentium/l33 PC and the experimental results have demonstrated its feasibility