Development of video retrieval engine by using a large-scale video corpus

2016 Fiscal Year Final Research Report

• Project/Area Number: 26280040
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Partial Multi-year Fund
• Section: 一般
• Research Field: Multimedia database
• Research Institution: Kobe University
• Principal Investigator: Uehara Kuniaki 神戸大学, システム情報学研究科, 教授 (60160206)
• Co-Investigator(Kenkyū-buntansha): 松原 崇 神戸大学, その他の研究科, 助教 (70756197)
• Research Collaborator: SHIRAHAMA Kimiaki ジーゲン大学, パターン認識グループ, ポスドク
• Project Period (FY): 2014-04-01 – 2017-03-31
• Keywords: 機械学習 / 情報検索 / 映像データ / コーパス / 映像検索

Outline of Final Research Achievements

We have developed a large-scale video retrieval method that can identify videos relevant to a semantically high-level query (e.g., “playing a guitar outdoors), by combing recognition results of concepts (abstracted names of meanings that a human can perceive from a video). Despite the recent advancement of concept detection, it is still difficult to accurately recognise various concepts. Thus, based on Dempster-Shafer Theory, we have quantised uncertainties in concept detection, and developed a probabilistic method that can perform accurate retrieval even using uncertain (erroneous) concept detection results. In addition, by adopting a visual attention model, we have developed a video retrieval method that reflects user’s intention by considering which concepts attract his/her attention.

Free Research Field

人工知能、特に機械学習