Improving Character Segmentation of Calligraphy Stone Rubbing by Adaptive Threshold

• Main Authors: Dong-Yan Hong, 洪東巖
• Other Authors: 廖俊睿
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/zrsvf9

碩士 === 國立中興大學 === 電機工程學系所 === 102 === The purpose of this thesis is to propose a segmentation system for Chinese calligraphy stone rubbing. The system can be divided into two parts: (1) Edge detection and (2) Adaptive threshold segmentation. For edge detection, we combine two different edge methods to increase the robustness of the detection. The first is Canny edge detection and the second is Otsu thresholding with constrast enhancement. Due to the different characteristics of the two edge detection techniques, anisotropic diffusion filters with different parameters are used to reduce the noises of the input images. For adaptive threshold segmentation, we use a block-based approach by setting different thresholds to each block according to the characteristics of the block. Since the most important part of the calligraphy stone rubbing is the brush stroke. In order to recognize a stroke in a block, the block size is determined by the stroke width so that a block can cover a stroke. Each block is then categorized as either a text block or a background block. The threshold is set to be higher for a text block and it is set to be lower for a background block. Within each block, each pixel is determined to be either a stroke pixel or background pixel according to the threshold. Finally, a post-processing step removes artifacts that are not recognized in the first two stages. Experiments show that our method is better than previous segmentation methods.