Character region identification from cover images using DTT

Abstract

A robust character region identification approach is proposed here to deal with cover images using a differential top-hat transformation (DTT). The DTT is derived from morphological top-hat transformation (TT), and efficient for feature identification. This research is considered as a fundamental study for auto-classification of printed documents for organizing a Digital Library (DL) system. The entire procedure can be divided into two steps: region classification and character region identification. In the first step, a source gray image is segmented by a series of structuring elements (SE) into sub-images using the DTT. Since the widths of regions are relative to the scales of the characters, the different scales of characters are classified into the series of sub-images. The character region identification processing is composed of feature emphasis, extraction of candidate character regions and region reconstruction processing. Feature emphasis processing reduces noises and emphasizes characters in the sub-images, and then the candidate character regions are extracted from the gray scale sub-images by a histogram analysis. Lastly, a morphological image reconstruction algorithm based on conditional dilation is introduced to make the extracted character regions distinct from noises. To demonstrate the robustness of the proposed approach, 30 gray scale cover images were tested in the experiments, which revealed that an average extraction rate of 94% has been achieved. © Springer-Verlag Berlin Heidelberg 2004.