Font Size: a A A

Application Research Of Ancient Book Image Annotation Technology Based On Density Peak Clustering

Posted on:2019-09-01Degree:MasterType:Thesis
Country:ChinaCandidate:W D GongFull Text:PDF
GTID:2438330551960572Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
The image of Ancient Books,as a medium that exists in the form of ancient texts,records the essence of human thinking about society and nature in the past.To meet the demand for labeled samples in normalized image processing of an-cient texts images,this paper mainly focuses on the density peak clustering anno-tation of Ancient Books.We carry out a research of Ancient Books clustering method based on information entropy improvement.And we optimize the cluster-ing algorithms of Ancient Books images by using correlation analysis.We have also designed an application system based on Ancient Books images annotation technologies.We include our main works as follows:(1)Our first work,given the fact that the Ancient Books processing field needs to use many labeled sample data,we studied the classification method of Ancient Books images based on density peak clustering,then we design a measure function of density peak clustering to improve the cluster analysis method for An-cient Books images based on information entropy.The method uses distance threshold enumeration to cluster,and to get the information entropy of the cluster-ing result,and by analyzing the attenuation of the information entropy we obtain the clustering threshold results,thus completing the clustering of the image.The improved metric function is used to measure the similarity between images and the greedy strategy,it is used as the basis for class member's merging operation to realize the increase of information entropy attenuation.Experiment on the dataset of the Yi Character shows that the method can classify the unknown segmentation images very well under non-super-large quantities,and then improve the effi-ciency of manual image collection and labeling.(2)Our second work,in the information entropy improved Ancient Books clustering algorithm,there is a misjudgment operation when evaluating and merg-ing through the threshold.Therefore,we propose an association analysis and op-timization method for image clustering.The method firstly uses the image cluster-ing set under density peak clustering to normalize the key point frequency matrix,then analyzes the frequent items of the matrix set and correlates and analyzes the frequency rules of the key points,and finally combines the similarity function when performing the member merging operation.Experiment shows that this method can provide significant help in image similarity judgment.(3)Our third work,to verify the information clustering algorithm based on improved density of information entropy and the method of correlation analysis to optimize the Ancient Books images,and to further perfect the application research work of the images annotation technology,we carry out the research and design work of the prototype system of the images annotation technology.The system can complete the management and display of labeling data and provide the anno-tation service of Ancient Books and the construction and management services of sample data sets.This helps to promote the identification and detection of ancient literature images.For our work,the theory and technology of Ancient Books images annotation proposed in this paper will help construct the Ancient Books images database,which can provide technical support for the identification,detection and retrieval of ancient documents.The exploration process of Ancient Books classification by clustering has certain reference significance.The work of this paper will also help promote the dissemination and protection of knowledge in Ancient Books and pro-vide theoretical and technical support for the Ancient Books' labeling field.
Keywords/Search Tags:Clustering optimization, Semi-supervised labeling, Ancient Books Image annotation, Density peak clustering, Archives of Ancient Books
PDF Full Text Request
Related items