Development of Efficient Video Search Methods
Sejong University announced on the 18th that a research team led by Professor Choi Yoo-kyung from the Department of AI Robotics has published a paper in AAAI (Association for Advancement of AI), the world's most prestigious artificial intelligence conference.
Hyunwoo Kim (from the left), Geuntaek Lim, and Kwangjin Lee, master's degree students in the Department of AI Robotics at Sejong University, are taking a commemorative photo. [Photo by Sejong University]
The paper, titled 'VVS: Video-to-Video Retrieval with Irrelevant Frame Suppression,' developed an accurate and efficient video retrieval method by suppressing unnecessary frame information in the video search process.
Previous descriptor research for video retrieval lacked consideration of distractor frames in raw videos, limiting advanced visual similarity calculations. The research team assumed that distractor frames have a strong correlation with the accuracy of video retrieval and conducted experiments by removing distractor frames.
The team proceeded with two stages of approach. First, they introduced the 'Easy Distractor Elimination Stage,' which removes distractor frames that are easily distinguishable based on visual information. Second, they proposed the 'Suppression Weight Generation Stage,' which suppresses distractor frames based on the video's topic and semantic correlation. This approach demonstrated the potential to overcome the limitations of existing video retrieval research and received high evaluation.
Im Geun-taek, a master's degree student, said, "It is an honor to present a paper at a globally recognized international AI conference. I want to continue growing as a researcher conducting valuable and important research."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
