| With the rapid development of computer technology and network technology,vast amounts of image information are digitized multimedia information, how toefficiently search the digitized image data has become an urgent problem to be solved.Content-based image retrieval technology is one of the most widely used solution, socontent-based image retrieval technology has become a hot topic in today’s computerfield. In this technique, different regions are selected and different image features areextracted in the retrieval system according to the different objects of study. And thusto feature region location and feature extraction is a key technology in the searchtechnique. The object of study is a precious Thangka in this paper, digital research onThangka is to accelerate the construction of China’s cultural industry, and also greatimportance to the protection of the cultural heritage of Thangka. In order to achieveefficient retrieval of Thangka image, this paper propose a method that is suitable formulti-granularity retrieval Thangka image after researching on content-based imageretrieval.Firstly, this paper systematically study the characteristics of Thangka imagecomposition, and find that the main statue and the headdress region possess thecharacteristic of circle. In the process of extracting headdress, headdress segmentationalgorithm for Thangka image based on circular region location is proposed, whichcombine with the characteristic of headdress region according to headdress locationbased on threshold value. The algorithm can extract the main statue region andsegment the headdress of the main statue. The experiments show that the headdressextraction algorithm of Thangka image can accurately locate headdress area and notincompletely segment headdress for the change of the main statue of proportion andthe size of headdress. Secondly, according to the main statue region, the headdressand the whole Thangka image, Thangka retrieval method based on multi-granularityis proposed, which combine with quotient space granular computing theory. With theproposed method, different quotient spaces are constituted according to differentbehaviors under different granularities of image, and the synthetic feature is obtainedby composing the attribute functions in different granularity levels based on the theory of composing multi-granularity attribute functions in the quotient space, theimages are finally retrieved images by the synthetic. Experimental results indicate thatthe proposed method is superior to the method based on single attribute, improvedlocal accumulate histogram-based Thangka image retrieval and the method based oncontour and color of Thangka headdress. |
