Design And Implementation Of Fabric Image Retrieval System

In the textile industry,the work of inquiring a picture of cloth flower case is usually done by the staff with personal experience.The query speed and query accuracy of this method are unstable,and can not satisfy the demand of companies.There is an image retrieval system for clothing retrieval in the market,but the particularity of the cloth flower case image makes the algorithm and system of these retrieval dress can not fully meet the retrieval needs of cloth flower case images.In this thesis,an image retrieval system is designed and implemented to retrieve the image of cloth flower case.The system is divided into two parts,the client and the server.The client includes request management module,data transmission module and picture display module,which are used to manage client's request,data transmission and display of similar images,respectively.The server includes request connection management module,data transmission module,search module,storage module and other functional modules,which are used to manage requests,data transfers,retrieve similar images and log management on server side respectively.In order to ensure the retrieval effect of cloth flower case image,we try to use the color moments,GIST,SIFT and other low dimensional features to characterize the fabric images,and select GIST and SIFT as feature extraction algorithms.Then we integrate GIST and SIFT,and presents an image retrieval algorithm based on SIFT comer feature and GIST spatial information feature,which is applied to retrieve cloth flower case images.In addition,in order to make the texture of the fabric image do not affect the feature extraction of SIFT,the algorithm of structure extraction from texture via relative total variation is introduced in this thesis.In order to get the most appropriate processing for the fabric images,we make an adaptive improvement to the algorithm of structure extraction from texture via relative total variation.The test shows that,when retrieving similar cloth flower case images,the accuracy rate of image retrieval algorithm of the system is about 80%and the recall rate of the same image is about 75%.The system can support 1000 queries at the same time,and the single retrieval time takes about 15 seconds.The system can meet the retrieval requirements of the cloth trading company.