Hierarchical content classification and script determination for automatic document image processing

Page segmentation and image content classification plays an important role in automatic document image processing with applications to mixed-type document image compression, form and check reading, and automatic mail sorting. In this paper, we propose an enhanced background-thinning based page segmentation algorithm to process document images rapidly and eliminate some small regions embedded in other regions. We then present a hierarchical approach, which combines cross correlation measure, Kolmogorov complexity measure, and a neural network, to classify sub-images into halftones and texts. The approach also achieves high accuracy in text determination using a three-layer feed-forward network, where text region can be classified into Chinese or alphabetic character. Experimental results on a number of mixed-type document images show the efficiency and effectiveness of our approach.