Optimization Of Image Retrieval Algorithms Under Complex Conditions And Realization Of Network Service With Image Retrieval

With the continuous development of Internet technology,image retrieval technology is more and more widely used in real life,especially in the fields of e-commerce.For example,on Taobao,Jingdong and other e-commerce platforms,user inputs the images of the products he wants to buy,and the platform uses image retrieval and other technologies to return the products that user wants to buy.This is a common application scenario of e-commerce platforms.The information contained in the image is more specific and simpler than the text or audio,the storage occupied by the image is smaller than the video,and the use of images is more convenient than the video.Therefore,image retrieval has always been a research hotspot on the Internet.In the wine industry,the production of wine bottles and the brewing of wine are usually separated.The buyer provides the image of the wine bottle to the manufacturer,and the manufacturer's staff search for the images of the database of wine bottle that they can produce according to the input image.If the staff of the manufacturer manually search and compare the image of the wine bottle,it will consume more labor costs of the manufacturer.According to the requirements of the manufacturers,this thesis implements a networked application of a wine bottle retrieval system.And it can provide access services for image retrieval of multiple users at the same time,and each bottle image retrieval can return the first 16 bottle images with the most similar shape of the input bottle image within 30 seconds.Based on the existing research on wine bottle retrieval,this thesis proposes related algorithms for the problems and characteristics of the wine bottle images to be retrieved,which proposes the preprocessing algorithm for image highlight removal and the semi-automatic image segmentation algorithm,and realizes the network application of the wine bottle retrieval service.The main work of this thesis are as follows:(1)Design and implementation of fusion bilateral filtering and Multi-Scale Retinex with Color Restoration algorithm.For the problems of highlights,shadows,and edge blur in the wine bottle image to be retrieved,this thesis proposes an algorithm for the highlight removal preprocessing of the bottle image to be retrieved,which fusion bilateral filtering and Multi-Scale Retinex with Color Restoration.(2)Design and Implementation of semi-automatic image segmentation algorithm.For the problem of the bottle image segmentation results,this thesis proposes a semi-automatic image segmentation algorithm,and use this method as a supplement to the automatic image segmentation method.The bottle image segmentation results may be over-segmented or under-segmented due to problems such as highlights,shadows,and blurred edges of the bottle.The semi-automatic segmentation method combines the region growing method and the semi-automatic segmentation idea.Before image segmentation,artificially mark the front scenic points or background points of the image to be retrieved,and use the marked information to constrain the segmentation process of the wine bottle image,and finally complete the wine bottle shape extraction of the wine bottle image to be retrieved.(3)Design and realization of the network application of the wine bottle retrieval system.The wine bottle retrieval system is implemented with a C/S structure,in which the server uses Spring Boot and other technologies to provide a Restful API interface,and the client is developed based on the Android system.Aiming at the network application of the wine bottle retrieval system,this thesis conducts comparative experiments on the highlight removal preprocessing algorithm and the semi-automatic image segmentation algorithm for the wine bottle images to be retrived.This thesis uses ten types of representative wine bottle images to be retrieved for experiments.The experimental results are as follows:(1)The highlight removal preprocessing algorithm in this thesis is superior to the existing algorithms in processing result and processing time efficiency.The algorithm in this thesis can not only remove the highlights and the shadows in the image,but also protect the outline of the wine bottle,at the same time,the preprocessing time has increased by nearly 8.4 times,and use the preprocessing algorithm in this article to preprocess the image and then the time required for automatic segmentation has increased by nearly 2.6 times(2)The highlight removal preprocessing algorithm in this thesis preprocesses the image and then uses the fully automatic segmentation method to segment it.On average,11.3 wine bottle images are segmented and retrieved correctly.The existing algorithm^[1]is used to preprocess the image On average,the retrieval results of 7 wine bottle images were correct,and the correct retrieval results improved about 4 wine bottle images.(3)The semi-automatic image segmentation method in this thesis is superior to the full-automatic image segmentation method in terms of segmentation result and segmentation time efficiency.The semi-automatic image segmentation method can roughly extract the outline of the wine bottle,and the image segmentation time has increased by nearly 1.6 times.