Study On The Correlation Fragments Retrieval In Dunhuang Manuscripts Conjugation And Its Implementation

The Dunhuang manuscripts are a cache of important religious and secular documents discovered in the Mogao Caves of Dunhuang,China,in the early 20th century.The Dunhuang manuscripts database project,which is a national key scientific research project,can make scholars research the Dunhuang manuscripts with browser easily.With time passed by,there are a lot of debris fragments in Dunhuang manuscripts,many of which can be conjugated.But it is hard to find out the correct fragments used in conjugation which is also named "ZhuiCan",as the result of the huge numbers of the Dunhuang manuscripts.With the development of digital technology,it is possible to join two parts of one Dunhuang manuscript together with digital image retrieval technology,and it is also the major research of this article.In this paper,the research works are as follows mainly;Firstly,as the demand of conjugating two parts of one Dunhuang manuscript,this paper identifies three major features of the Dunhuang manuscript images,including the texture feature,the edge feature and the font feature.Present a feature model of the Dunhuang manuscript images.For the texture feature,study the feature of color of the Dunhuang manuscript images.Design the method to filter the major color,namely color of texture,and minor color,namely the color of background and words,for the color distribution of image is very uniform.Design the texture histogram,which is based on major color histogram.Define the texture histogram as the texture feature.For the edge feature,due to the main consideration of "ZhuiCan" is the matching degree of left and right edges,present a left and right edges detection algorithm,which is based on the Canny edge detection,and define the points of left and right edges as the edge feature.For the font feature,study and improve the multispeed clustering algorithm based on max-min distance means,design a method to detect and separate the feature points of each word,which based on SURF and max-min distance means.Define these feature points of each word as the font feature.Secondly,study and define the un-similarities of each aspects of the feature model of the Dunhuang manuscript images.Present the un-similarity of the Dunhuang manuscript images and then a correlation fragment detection algorithm based on this un-similarity.Study the EMD distance and define the texture un-similarity based on it.Design a method to calculate the offsets of left and right edges.Define the edge un-similarity based on Hausdorff distance without offsets.Study the method to build orientation vector histogram by using feature points of each word.Define the font un-similarity based on the EMD distance of orientation vector histogram.Present the un-similarity of the Dunhuang manuscript images based on un-similarities of each features.Present a correlation fragment detection algorithm based on the un-similarity.First calculating the clustering of all Dunhuang manuscript images then calculating the un-similarities of each two image in every clustering.Finally,in order to meet the demand of Dunhuang manuscripts database project,design the organizational structure of second-phase of this project.Implement main functions of it.On the basis of the algorithm mentioned before,implement the correlation fragments browse functions.