The Design And Implementation Of Image Retrieval System For National Pattern Image

With the progress of science and technology,the rapid development of modernization,people's lifestyle and production mode has changed.Traditional culture and traditional technology is the product of the historical precipitation,which is also an indispensable precious resource.The protection of national cultural heritage has become more and more important.National pattern takes an important part in traditional cultural heritage,the majority of which have a specific meaning or connotation.To extract national pattern from the image and establish a national pattern database is of great significance for the protection and inheritance of cultural heritage and inheritance.Meanwhile,combined with the image processing technology to develop a retrieval tool for the national pattern,whether in the retrieval and recognition area,data mining and analysis,or in the auxiliary design of commercial production is of great value.This project comes from Beijing Municipal Science and Technology Commission,the project of "The traditional culture pattern feature extraction and the application of costume design".The main purpose is to extract the national pattern,which is used in the costume design combined with image retrieval technology.Image retrieval has been widely used in various fields.Based on the requirement of extracting the national pattern to establish a pattern database and the needs of practical retrieval application,this thesis has designed and realized an image retrieval system for the national pattern.The main works of this thesis are as follows:(1)Firstly,this thesis studies the relevant theories and key techniques of content-based image retrieval.Analyze the characteristics of the national pattern image.In view of the specific functional requirements of the system in image segmentation and image retrieval,this thesis regards image segmentation and feature matching algorithm as research emphasis.Making comparison and analysis the applicability of different algorithms in the national pattern and the effect of treatment.Based on the comparison of results,chose EGBIS as the image segmentation algorithm,SURF and shape context as the feature matching algorithm for system implementation.(2)In order to reduce the image matching times and improve the efficiency of the national pattern retrieval,this thesis leans from the idea of hierarchical clustering and builds a hierarchical retrieval model.In the classification layer,aiming at extract the invariant moments as global features.It also proposes a normalization calculating method for the invariant moments.Utilize the Mahalanobis distance to replace Euclidean distance as the similarity measure in the clustering.Hence make it more applicable to the national pattern.In the retrieval layer,in order to ensure the recall of the system,it executes the precise matching both in the nearest cluster and the second nearest cluster.The result of evaluating test indicates that the proposed hierarchical retrieval method greatly reduces the number of matching times and improves the system efficiency.(3)This thesis has designed and made the implementation of a national pattern retrieval system according to the functional requirements.In order to facilitate the correlation information of the images,it defines original image,sub image and super image.Relational database is established depending on these definitions.The national pattern is divided into three levels.We execute the searching schema on the super image level.The query of original image is achieved by index.Finally,the national pattern retrieval system is implemented,consists of automatic national pattern segmentation,database management,national pattern retrieval,original image searching and the module for viewing image structure.