The Design And Implementation Of Content Based Image Retrieval System

The research topic of this thesis is the design and implementation of a content-based image retrieval system(CBIR),which comes from the internship project of NetEase Games Interactive Entertainment Division.The system is aimed to quickly retrieve image from a large number of game libraries and is becoming an indispensable tool for the developers and arts personnel,in order to improve the efficiency of the game department.This thesis starts from the requirement analysis and discusses the principle and technology of CBIR.The architecture of this system is divided into two parts:real-time and offline,and they are refined into five components,namely preprocessor,feature extractor,indexer,retriever and graphical user interface Components.Then,we elaborate the main principles and design ideas of each component,and give the specific implementation,Finally,we show the data results.The main part of the offline part of the system is to create a feature index library in advance to improve the performance of feature vector matching.The real-time part of the system is mainly for users to submit images,the system retrieves images and displays the search results.The main purpose of the preprocessor is to denoise the retrieved image and reduce the scale difference.The feature extractor implements multi-feature extraction based on color,shape structure and neural network.The indexer function creates a feature index library offline for the image library,and the searcher completes feature matching.Sort the results according to the similarity,and finally display the search results through the graphical user interface.The preprocessor adopts a generative anti-network(GAN)to make image denoising,which has achieved good results in text noise and background noise processing.The searcher uses a K-d tree structure to improve the retrieval speed performance.The thesis focuses on the design of system components and the implementation of related algorithms.