Vocabulary Tree-based Image Retrieval, Extraction And Research Of Visual Keywords

Speed and precision are the two basic requirements for image recognition and retrieval algorithm. Content-Based Image Retrieval (CBIR) technology has made great progress for text retrieval technology, it usually take the rank algorithm, but the retrieval method on efficiency have apparently can't meet the needs of users for searching of large capacity, high dimension image.Vocabulary tree is a high efficient data structure for image retrieval based on the visual keywords than other structure. Expansibility and adaptability are easy to get satisfaction using the vocabulary tree in image retrieval algorithm, so as to realize the large-scale image retrieval task. Based on the vocabulary tree is a pretty big system in large-scale image retrieval. So our team members accomplish the system together. In this thesis, I emphatically research the retrieval method for visual keywords in the team, and also I study an algorithm for image retrieval based on vocabulary tree.The key technology for generated the visual keywords is the extraction of the visual information in image. This paper use the extraction method for visual keywords is integrated the multi-features of the region of the interest points. We adopt the Harris algorithm to detect the interest points, and using this method to determine the region of the points, and we use the multi-feathers on the region of the interests, such as colors and textures to extract the feathers of images in data base. Due to the feather vector's dimension is higher, so we adopt the improved k-means algorithm to cluster these feather vectors, and clustering center as a visual keyword after clustering.Feature vectors of every image should be quantified to visual key words, so as to don't consider the position of the feature vectors in a image, those visual keywords will be as leaf nodes in a vocabulary tree, then using vocabulary tree method to retrieval image for large-scale image retrieval.