Research On Character Extraction And Segmentation Of Chinese Historical Seal Images Based On Multiple Features

Today,with the increasing popularity of digital image processing technology,combining ancient historical resources with cutting-edge technologies to realize its digitization is an important way to protect cultural heritage.As an important part of cultural heritage,book paintings cannot be ignored for their protection and research.Chinese historical seals often appear in a series of cultural heritages such as book paintings to identify author information.In addition,the characters in the Chinese historical seal are an important basis for the management of these cultural heritages.In order to protect and further study this cultural heritage,the character extraction and the character segmentation of seal images are a necessary step.However,due to the scarcity of Chinese historical seals and the variety of carving methods,the character extraction and the character segmentation of Chinese historical seal images is challenging.The thesis mainly studies the character extraction and segmentation of Chinese historical seal images.First,the multi-structural features of the seal image are constructed.Then the decision fusion method based on multi-features and decision trees is designed to achieve classification of seal carving.Finally,according to the classification result,the adaptive region extraction method is designed to extract the characters of seal images,and the single-character segmentation is realized based on the morphological analysis.The main research contents are as follows:1.Three statistical features for the carving method of Chinese historical seal images are constructed,and a seal classification method based on decision fusion is proposed.The border statistics feature and the character distribution feature of the seal image are extracted by analyzing the probability distribution of border width and column brightness change of the seal image,and the decision tree based on the border statistics feature and the character distribution feature is generated to realize the seal carving classification(rilievi and diaglyph).By analyzing the brightness change of the frame structure decreasing by pixel starting from the outer contour of the seal image,the border-level feature is extracted.Based on the border-level feature,the fine classification of the diaglyph images is achieved.2.A single character segmentation method based on morphological analysis is proposed to realize character extraction and single character segmentation.Accordingto the carving method category of the seal image,the category-aware binarization and the adaptive character region extraction are performed,and the single-character segmentation of the seal character image is realized by morphological analysis.3.Based on the above seal classification and seal character segmentation,a framework of multi-features based character extraction and segmentation system for Chinese historical seal image was designed.In Matlab,the corresponding system is developed by using GUI tools,which realizes the functions of seal carving classification,character extraction and single character segmentation.In this paper,the validity of the proposed method is verified by experiments on the real data set of Chinese historical seal images.The experimental results show that the proposed multi-features based character extraction and segmentation method for the Chinese historical seal images can classify carving methods of seal images accurately,extract characters effectively and segment characters accurately.