Divide-and-conquer approach to Japanese text segmentation

Stephen W. Lam; Qunfeng Liao; Sargur N. Srihari

doi:10.1117/12.205824

Translator Disclaimer

30 March 1995 Divide-and-conquer approach to Japanese text segmentation

Stephen W. Lam, Qunfeng Liao, Sargur N. Srihari

Author Affiliations +

Proceedings Volume 2422, Document Recognition II; (1995) https://doi.org/10.1117/12.205824
Event: IS&T/SPIE's Symposium on Electronic Imaging: Science and Technology, 1995, San Jose, CA, United States

Abstract

This paper presents a robust text segmentation algorithm for printed Japanese documents. A divide-and-conquer approach is proposed to handle a large variety of image qualities and print styles. The approach can adapt its processing strategies according to the text quality, i.e., a method using diverse knowledge sources will be exploited to segment degraded text while a fast simple method will be used for good quality text. Since the algorithm can adaptively select the methods for different scenarios, the segmentation is highly efficient in terms of speed and accuracy. The segmenter has tree modules for image preprocessing, line segmentation, and character segmentation. The preprocessor uses the statistical information of the image connected components to globally estimate character size and uses projection profile to determine image quality. The line segmenter requires a `thresholding and smoothing' step prior to line extraction if the image is noisy. During character segmentation, the character segmenter first tries to locate components which contain touching characters. If touching characters exist, an algorithm which includes a profile-based splitting and classifier-based multiple hypothesis processing will be invoked to perform the segmentation.

Citation Download Citation

Stephen W. Lam, Qunfeng Liao, and Sargur N. Srihari "Divide-and-conquer approach to Japanese text segmentation", Proc. SPIE 2422, Document Recognition II, (30 March 1995); https://doi.org/10.1117/12.205824

ACCESS THE FULL ARTICLE