Research On Fuzzy Retrieval Method Of Chinese Character Images In Ancient Books Based On Multi Attributes

Image retrieval of ancient Chinese character is an effective way for researchers to obtain associated ancient Chinese glyphs.Due to the large number,complex structure,and variable shape of ancient Chinese characters,however,it is difficult for traditional Chinese character image retrieval technology to obtain satisfactory results.Therefore,it is necessary to develop an effective image retrieval technology of ancient Chinese characters to meet the practical needs based on the characteristics of ancient Chinese characters.The key technique of image retrieval of ancient Chinese characters is developed through analyzing and summarizing the characteristics of ancient Chinese characters,and introducing a hesitant fuzzy set theory.The main work is divided into the following two parts:(1)Segmentation algorithm of ancient Chinese character images based on hesitant fuzzy set is proposedThe algorithm of image denoising and segmentation is designed to get the initial segmentation results of ancient Chinese characters.On this basis,the over and under segmentation errors in the initial segmentation results are corrected.The hesitation fuzzy set is established based on the advantages in the aspect of dealing with multi-attribute decisionmaking problems,and realize the recognition and merging of regions with over-segmentation errors,then the segmented pixel jump number mutation analysis method is used to segment Chinese character regions with adhesion and overlap problems to obtain single-word images of ancient Chinese characters.In this study,92 sample images(28886 words)of Wen Yuan Ge,Wen Jin Ge,Wen Shuo Ge and Wen Lan Ge of Si Ku Quan Shu are used to test.Results showed that merging accuracy rate of segmented Chinese characters is 85.7%,and the accuracy of the Chinese characters is 92.3%.(2)Image retrieval algorithm of ancient Chinese character based on hesitant fuzzy weighted distance measurement is proposedOverlapping normalized bi-elastic grid division method is used to the feature selection and extraction of ancient Chinese character image,and then summarized the multi-dimensional retrieval attribute features of ancient Chinese character image.The hesitant fuzzy sets of image retrieval of ancient Chinese characters were constructed based on the defined membership function corresponding to each index under the stroke feature,corner feature,font structure feature and statistical feature of ancient Chinese character image.Finally,the fuzzy weighting distance measure between the target image and the image to be retrieved is used as the similarity measure to obtain the retrieval results of ancient Chinese characters.The result showed the recall and precision of the retrieval experiments on the segmented 26661 ancient Chinese character images are 76.5% and 78.9%,respectively.These results revealed that the proposed Chinese character image retrieval method of ancient books can adapt to the characteristics of ancient Chinese character image and achieve higher performance.