Research On Feature Extraction And Evolution Patterns Of Auspicious Cloud In The Dunhuang Grotto Murals

The Dunhuang grotto murals are the fusion of the traditional Chinese religious art,painting art and literature art. They are the main part of Dunhuang art. Theauspicious cloud which reflects the flying status of the flying apsaras is the mostwidely used element in Dunhuang grotto murals. As time goes on, the auspiciouscloud shows the evolution patterns. However, the current research of the auspiciouscloud focuses on using priori knowledge from macro aspect to analyze theircharacteristics and evolution patterns. Its result is described abstractly, and lacks of aprecise quantitative analysis. Thus, it’s significant to explore Dunhuang art byextracting features of the auspicious cloud and analyzing quantitatively with thelatest image processing technology.To extract the auspicious cloud’s features and analyze the evolution patterns, theresearch work of this thesis is shown as follows:1) According to the characteristics of the auspicious cloud in Dunhuang grottomurals at different times, appropriate pre-processing steps were applied to providebasic data for further extracting features. Combining with the history of theauspicious cloud and its macro form, the features of the auspicious cloud in eachperiod were analyzed. Basic data was obtained for further feature extraction byseveral pre-processing steps, such as converting the image color space, graying,binarizing and extracting contours.2) The feature values of color, texture and shape, which were extracted toquantitatively describe the auspicious cloud, were fundamental to research evolutionpatterns of the auspicious cloud. The class number feature of color and colordistribution feature were calculated by using color histogram and color momentrespectively. Texture features were extracted from both spatial and frequencydomains. For spatial domain, the correlation features of gray-level space were fromstatistical analysis. For frequency domain, texture features were obtained by usingGabor wavelet transform. The shape features based on central moments and thesmallest enclosing rectangle were extracted to describe the overall shape, and theconcavity and convexity property of the auspicious cloud’s contour were obtained by curvature scale image. The head area and tail area from the cloud weresegmented by using the maximum inscribed circle method, and then features of thetwo local regions were extracted.3) The evolution patterns of the auspicious cloud in eleven periods were analyzedby using statistical analysis method, and discrimination of the auspicious cloud’speriod was implemented. The quantitative feature values of the cloud in elevenperiods were extracted to statistically summary its evolution patterns. Then thefeature extraction methods were applied on detecting which period the auspiciouscloud was in, and the feature scope shrinking detection algorithm was proposed.Recall ratio, precision ratio and accuracy ratio were used to describe performance ofthe algorithm. The experimental results showed that the proposed features haddiscrimination about the auspicious cloud in eleven periods.