講演抄録/キーワード |
講演名 |
2015-09-04 10:15
Nested Lattice Hashing Scheme for Similarity Search Applications ○Thanh Xuan Nguyen・Ricardo Antonio Parrao Hernandez・Brian Michael Kurkoski(JAIST) IT2015-37 |
抄録 |
(和) |
This research targets improving similarity search efficiency using a nested lattice hashing scheme. Similarity search has been established as a fundamental paradigm for a variety of applications, including information retrieval, data mining, multimedia database searching and machine learning. In principle, the similarity search problem is to find the object (e.g., image, sound, video, file) most similar to a given object in a set of objects, which are usually represented by a collection of real number feature vectors in Euclidean space. On the other hand, lattices form effective structures for various geometric, coding and quantization problems. This research takes advantage of lattices for quantizing feature vectors to hash values. The goal is to develop a lattice-based hashing scheme such that there is a proportional relationship between Euclidean distance and a metric we call first difference distance between features vectors. The proposed two-dimensional nested lattice code reduces the normalized mean square error (NMSE) by 20% compared to two-dimensional Gray code. |
(英) |
This research targets improving similarity search efficiency using a nested lattice hashing scheme. Similarity search has been established as a fundamental paradigm for a variety of applications, including information retrieval, data mining, multimedia database searching and machine learning. In principle, the similarity search problem is to find the object (e.g., image, sound, video, file) most similar to a given object in a set of objects, which are usually represented by a collection of real number feature vectors in Euclidean space. On the other hand, lattices form effective structures for various geometric, coding and quantization problems. This research takes advantage of lattices for quantizing feature vectors to hash values. The goal is to develop a lattice-based hashing scheme such that there is a proportional relationship between Euclidean distance and a metric we call first difference distance between features vectors. The proposed two-dimensional nested lattice code reduces the normalized mean square error (NMSE) by 20% compared to two-dimensional Gray code. |
キーワード |
(和) |
Nested lattice / hashing scheme / similarity search / first difference distance / / / / |
(英) |
Nested lattice / hashing scheme / similarity search / first difference distance / / / / |
文献情報 |
信学技報, vol. 115, no. 214, IT2015-37, pp. 19-24, 2015年9月. |
資料番号 |
IT2015-37 |
発行日 |
2015-08-28 (IT) |
ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
IT2015-37 |
|