Research On Fast Image Retrieval Algorithm Based On Visual Feature

In recent years,with the improving of the 4G network performance and the popularity of Wi-Fi,people can make use of mobile phone or tablet to shoot at any time and share their own image,computer vision technology is used to retrieve or recognition of these images,analyze the user behavior,provide data support for the from all walks of life.So the image retrieval identification technology is of great application value.In this paper,first of all,for millions of massive image database done in-depth research,the image retrieval based on visual vocabulary tree model engine is presented.Local visual features of image were analyzed and the extraction,the extraction of feature vector clustering training.By using the method of TD-IDF generates visual vocabulary tree index file,this file could achieve millions level image data retrieval in a second by single server process.Then put forward a kind of new method to describe image quickly.This method uses millions of images to build a vocabulary tree,and then extract the data from the vocabulary tree as a new visual descriptor,to complete the training for the new image of expression.By using a new method of describing vector to establish the visual query index,complete image retrieval functions.This method quickly produces a description relative to the traditional visual descriptor,being a small memory space,will also be able to satisfy the query image is a certain degree of scaling,rotation and scale can greatly shorten the time spent on training images,improve the efficiency of training images.The method of visual description vector in retrieval can also keep high retrieval accuracy.Then,in view of the database update problem,this paper proposes a incremental database image retrieval algorithm.Expansion and the method is used for database query index file for updates,for new image database,the extracted files on visual feature vector of the index tree traversal,through to the leaf node in the inverted file update weights,completed the expansion of the database updates.This method is compared with the traditional way of updating index files can greatly reduce the time to update the database,for applications requiring online database construction has important significance.At the end of the paper,we built a mobile terminal intelligent navigation system of the scenic spot based on the research of image search engine algorithm.The framework of the system are introduced,in detail elaborated the function of each module.The system has changed the traditional way people scan qr code for the scenic area information,by using a mobile phone camera to scenic area exhibition board scanned directly,return relevant attractions of geography,an introduction,travel guides,route planning and panoramic map navigation and so on.Improved people's travel experience has the very high commercial value.