Recognizing persons in television programs using region-based image features

Abstract

Technology that can search efficiently for a desired scene is essential for the management of large-scale TV program archives. A particularly common scenario of this sort is searching TV programs for scenes containing a specific person. There have been many previous studies on recognizing people in video footages. However, the recognition accuracy of their techniques drops significantly when people in the video appear under different lighting conditions, look in different directions, or have different facial expressions. In this paper, we propose a robust technique for recognizing persons in video, especially in television programs. Unlike the feature-point-based approach of the previous techniques, our method uses a new region-based approach inspired by general object recognition. Since it does not require facial feature points to be detected, it improves the robustness of the recognition to image fluctuations that are often found in television programs. In experiments, we verified the effectiveness of our method by using it to conduct searches on approximately 150 TV drama episodes. We also demonstrated the superiority of our method in comparative tests with conventional methods. © 2018 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.