A Fast Quantization Tree Based Image Retrieval Method

Traditional content-based image retrieval technology expresses the content of each image by feature vectors. Then image retrieval process begins by calculating the similarity between the image to search and the images in the database in terms of the corresponding feature vectors, next ranks the images in the database by a descending order of similarity, and finally outputs the desired top ones. This method shows good accuracy and high efficiency when the image databases are not very big. For modern large image databases, however, these methods will not satisfy users’ requirements for retrieval time and accuracy. To meet these challenges, in this paper, a new content-based image retrieval method is proposed which is similar to the Scalable Vocabulary Tree image retrieval method but with important variations. Experimental results show our method can more efficiently deal with larger image databases while with similar retrieval accuracy.