We propose GIFTS (Goods Image Features for Tree Search) which uses image local features for large-scale object recognition. Each GIFTS is a kind of keypoint feature. The feature vector consists of intensity deltas for 128 selected pixel pairs around the keypoint. By generating a KD-Tree from the GIFTS feature vectors of the training images and using the KD-Tree to search for nearest neighbor feature vectors of a query image, query times are on the order of log N for specific object recognition. We used the proposed method for book cover queries with 100,000 training images, had recognition accuracy over 99% with query times within one second.
CITATION STYLE
Nakano, H., Mori, Y., Morita, C., & Nagai, S. (2015). Large scale specific object recognition by using GIFTS image feature. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9280, pp. 36–45). Springer Verlag. https://doi.org/10.1007/978-3-319-23234-8_4
Mendeley helps you to discover research relevant for your work.