Japanese / English

Detail of Publication

Text Language Japanese
Authors Kazutaka Takeda, Koichi Kise, Masakazu Iwamura
Title Memory Reduction and Stability Improvement for Real-Time Document Image Retrieval with a Database of 10 Million Pages
Journal IEICE Technical Report
Vol. 110
No. 467
Presentation number PRMU2010-275
Location 茨城県つくば市
Reviewed or not Not reviewed
Month & Year March 2011
Abstract In this paper, we propose a large-scale document image retrieval method which realizes real-time processing with Locally Likely Arrangement Hashing (LLAH). Although LLAH has high accuracy and robustness, it requires a large amount of memory. It is also required to increase the discrimination power and stability of features for scaling up the database. For these purposes, we introduce the following three improvements. The first one is reduction of the required amount of memory by sampling feature points stored in the database. The second improvement is to increase the discrimination power by increasing the dimension of features. The last one is advancement of stability by removing redundant dimensions of features. From the experimental results, we have confirmed that the proposed improvements help to realize accuracy of 99.4% and processing time of 38ms for the database of 10 million pages.
Back to list