Research On Yan's Calligraphy Style Recognition Based On Features Fusion

The study of calligraphy style using iconography is of great significance in the application of calligraphy style recognition,classification storage and style appreciation.In the various Tang dynasty regular script,Yan Zhenqing's regular script is the most representative.Therefore,it is of great academic and practical significance to study the style of Yan's regular script in computer vision.This subject mainly studied the recognition of calligraphy style and the consistency of Yan's calligraphy style.We have classified Yan's calligraphy style into subcategories by identifying Yan's calligraphy style in different periods.In addition,we have verified the internal similarity of Yan's calligraphy style by identifying the font style of Yan's calligraphy and that of the other three calligraphers,so that we could distinguish Yan's style from other styles.The research content of this proj ect includes the following aspects:Firstly,the calligraphy font library was extracted as the datasets:In the process of getting the datasets,we first preprocessed the ancient inscription,the next,the preprocessed ancient inscription was denoised using run-length probability statistics,after denoising,the words in the inscription were cut separately by adaptive minimum bounding box cutting algorithm.Finally,the original datasets were obtained,which include four tablets of calligrapher Yan Zhenqing and the different calligraphic inscriptions of four regular script calligraphers.Secondly,the effectiveness of traditional image operators in calligraphy style recognition were studied.In terms of feature extraction,Gabor,wavelet,and LBP were studied,and the SVM classifier was used as recognition method.In view of the unique stroke features and structural features of calligraphy characters,we have selected effective direction and scale parameters when configuring Gabor filter.In this way,the information of calligraphy characters could be fully extracted and the redundancy of data could be reduced as much as possibleThirdly,a method based on the fusion of GIST and PHOG feature was proposed.We used GIST and PHOG to extract the global features and the local contour features of the calligraphy image respectively,then fed the two features to KPCA for dimensionality reduction,and connected the reduced features to form a fusion feature.Finally,the SVM classifier was used to perform the style training and classification.The experimental results proved the superiority of the proposed method and the specificity and representativeness of Yan's calligraphy styleIn conclusion,the calligraphy style classification method based on the fusion of GIST and PHOG could overcome the limitation of single feature in the description of calligraphy character features effectively.At the same time,the dimensionality reduction algorithm was applied to the fusion of calligraphy features,which could not only retained the rich features of calligraphy characters,but also reduced the redundancy of data and improved the recognition efficiency greatly.Furthermore,this subject played an important supplementary role in the study of calligraphy.