Research On Image Retrieval Algorithm Based On Bag-of-Words And Vocabulary Tree

With the development of computer networks and multimedia technology, large amounts of digital image resources are produced every day, effective management and efficient image retrieval algorithms become a research focus. The traditional text-based image retrieval realized simply, but has certain limitations because of time-consuming in manual tagging and subjectivity. Content-based image retrieval technology has become the mainstream of image retrieval in the last decade combines with computer vision, machine learning and information retrieval technology, and has broad application prospects in many fields.Based on the analysis of the status research on CBIR algorithms, we research on the image features extraction, high-dimensional indexing of massive data and sorting algorithm, and achieve image retrieval algorithm suitable for the different size of database. The goal is to improve the performance of algorithm and complete the design and implementation of prototype system of smart city street image retrieval.The main studies and results of the thesis are listed as following:1. Realize the image retrieval algorithm based on the Bag-of-Words model and SVM classifier. We introduced the SIFT and SURF local image feature extraction algorithm and described the bag-of-words models and the principle of support vector machine specifically. We also test the speed and retrieval performances on the different size of database using SIFT and SURF, then give the appropriate conclusions.2. Achieve the image retrieval algorithm based on the vocabulary tree. We study on the vocabulary tree algorithm, the sort algorithm and large-scale database memory strategies. Using KD-tree with BBF and RANSAC to remove false match points, we can improve the accuracy effectively. On the basis of these studies, fifty thousand images retrieval within seconds is realized.3. Achieve the prototype system of smart city street image retrieval. To meet needs, the system is designed to Client/Server architecture with front and rear ends. Client side communicates via Wi-Fi wireless network with server side. We introduced the function of system and the implementation of specific module. We tested the system to prove the feasibility of the system.