Japanese / English

Detail of Publication

Text Language Japanese
Authors Yuzuko Utsumi, Yuji Sakano, Keisuke Maekawa, Masakazu Iwamura, and Koichi Kise
Title Fast Face Recognition Based on Local Features and Voting Scheme
Journal Trans. IEICE
Vol. J97-D
No. 8
Pages pp.1263-1272
Reviewed or not Reviewed
Month & Year August 2014
Abstract Personal authentication for criminal investigations and immigration control requires finding a particular person from massive face images. The most important elements of the personal authentication are computational time and accuracy of recognition when massive images are registered in a face recognition system, namely scalability against the size of registered information. Many conventional face recognition methods can recognize face images accurately. However, if the number of registered images in- creases, computational time for recognition increases in proportion to the number of registered images and it is impossible to execute face recognition in real-time. This is because the conventional methods did not take the scalability into account. If we narrow the candidates of recognition down from registered images and apply the conventional methods to the narrowed candidates, we can realize a fast and high-accuracy face recognition system. In this paper, we propose a fast face recognition method which narrow down the candidates from massive face images by using local features, an approximate nearest neighbor method and a voting scheme. We made a 3 million face database and used the database as learning samples to evaluate the proposed method. The proposed method narrowed down recognition candidates to 1000 images with 257 ms and 96.9% accuracy.
URL http://search.ieice.org/bin/summary.php?id=j97-d_8_1263
Back to list