Conference paper
Using local maxima profile and Piece-Wise technique for line segmentation on Thai handwritten historical documents
2011 International Conference on Machine Learning and Cybernetics, pp.1862-1866
IEEE
International Conference on Machine Learning and Cybernetics, ICMLC 2011 (Guilin, China, 10/07/2011–13/07/2011)
2011
Abstract
This paper presents a new approach for segmenting text lines on Thai handwritten documents. The proposed technique is based on an Adaptive Local Connectivity Map concept using Piece-Wise Separating Lines. The algorithm is designed to solve problems in handwritten documents such as fluctuating text lines. Moreover, local maxima projection profile is used for enhancing the speed of extraction. The proposed algorithm consists of four steps. Firstly, Otsu algorithm is used to binarize the source image. Second, Piece-Wise Separating Lines is applied to derive the Adaptive Local Connectivity Map to show mask text lines. In the third step, local maxima projection profile is used as a guideline for extracting text lines. Finally, contour algorithm is used to identify the interested mask text line. The interested mask text is used to map with text image in order to extract the text lines. Analysis of experimental results on the King Rama 5 archive data indicated that the method has achieved a correct rate of 85.7%.
Details
- Title
- Using local maxima profile and Piece-Wise technique for line segmentation on Thai handwritten historical documents
- Authors/Creators
- S. Sangsawad (Author/Creator) - Murdoch UniversityR. Chamchong (Author/Creator) - Murdoch UniversityC.C. Fung (Author/Creator) - Murdoch University
- Publication Details
- 2011 International Conference on Machine Learning and Cybernetics, pp.1862-1866
- Conference
- International Conference on Machine Learning and Cybernetics, ICMLC 2011 (Guilin, China, 10/07/2011–13/07/2011)
- Publisher
- IEEE
- Identifiers
- 991005542028707891
- Copyright
- © 2011 IEEE
- Murdoch Affiliation
- School of Information Technology
- Language
- English
- Resource Type
- Conference paper
Metrics
222 File views/ downloads
75 Record Views