Face Indexing for Social Analysis

2014 Fiscal Year Final Research Report

• Project/Area Number: 24500136
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Media informatics/Database
• Research Institution: National Institute of Informatics
• Principal Investigator: LE DINH DUY 国立情報学研究所, コンテンツ科学研究系, 准教授 (80450170)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: 顔照合 / 顔インデキシング / 映像マイニング / 映像インデキシング / 社会分析 / クラスタリング

Outline of Final Research Achievements

We have developed advanced techniques for face indexing in large video archives with the target application being social analysis. As a result, we have improved performance of face detection, matching, and identification by taking the benefits from a long-term observation of large multi-channel video archives. We have developed a scalable technique for face retrieval and identification from large datasets of hundreds of millions of faces.
We have applied these techniques to the video archive of more than 100,000 hours of video programs broadcast on 7 channels in the Tokyo area. Then we have demonstrated an application for social analysis based on frequent appearance of politicians and celebrities.

Free Research Field

マルチメディア分析