Research Of Ancient Chinese Character Image Clustering Based On BIRCH And Improved K-medoids Algorithm

The Chinese characters are an important way for recording five thousand years ofChinese civilization. It is important to study ancient characters for studying Chinese culture.However there are many similar Chinese characters and large numbers of ancient Chinesecharacters with complex structure. These characteristics of the ancient Chinese charactersmake the study difficult. The clustering of ancient Chinese characters will bring a greatconvenience for the study. Based on the BIRCH clustering algorithm and improvedk-medoids clustering algorithm, an image clustering method using combined features isproposed.The ancient Chinese character images have poor print quality, different papers andtypesetting of the ancient Chinese words have some adhesion and dividing lines, these makethe deal with the ancient Chinese characters more difficult. we have to do somepreprocessing works to the ancient characters images firstly, because there are manydifficulties. For example, poor print quality, Chinese characters adhesion and moreunderscores. Then we extract the crude peripheral characteristics and the edge of thedirectional line element features of the strokes from four directions with the use of gridstructure. The combination of the two characteristics is the ancient character images’characteristics. In this paper, we design a new clustering algorithm for the ancient Chinesecharacter based on the BIRCH clustering algorithm and improved k-medoids clusteringalgorithm. The clustering algorithm is proved effective for the ancient Chinese characters byexperiments.