BROWSE
Detail
Conference Paper
Super-Node Selection Method for Incremental Vocabulary Tree
Cited - time in 
Share 
Share
- Authors
- Ho-Yong Seo, Wook-Ho Son, Ju-Jang Lee
- Issue Date
- 2012-08
- Citation
- International conference on humanized systems (ICHS) 2012, pp.1-5
- Language
- English
- Type
- Conference Paper
- Abstract
- Among the methods of object recognition, analysis based on image similarity has wide variety of applications such as image classification in robotics, image retrieval in search engine. Generally, image similarity requires large number of image database which had been classified. When a new query image is inserted to database, by analyzing image similarity of query feature, we can get the class or similar images of the query image by calculating term frequency-inverse document frequency (TF-IDF) score. Vocabulary tree (VT) is one of fixed size image database based on local feature extraction and image similarity. This structure has short similarity check time, but doesn't make any update for query image result information. To solve this problem, incremental vocabulary tree (IVT) structure which can perform self-update is proposed. But, each leaf node of IVT has varying number of images and features from self-update process. So, for the object of improving accuracy performance, proper node selection algorithm has to be defined. In this paper, when IVT structure is updated, super node which can make critical effect to accuracy performance will be chosen. So we can get fixed-size histogram vector by fixed-size super node, and also image retrieval performed by super node returns better accuracy. The additional time complexity could be negligible.
- KSP Keywords
- Accuracy performance, Image Classification, Image retrieval, Improving accuracy, Leaf node, Local feature extraction, Object recognition, Search Engine, Selection method, Time Complexity, Vocabulary Tree