Artificial Intelligence Major Professor Noh Junhyuk’s Research Team Wins Second Place at the International BlEmoRe Chall N
- Date2026.07.13
- 387
A research team led by Professor Noh Junhyuk of the Department of Artificial Intelligence won second place in the international BlEmoRe (Blended Emotion Recognition) Challenge, held as part of IEEE FG 2026 (International Conference on Automatic Face and Gesture Recognition), one of the world's leading conferences on face and gesture recognition.
Supervised by Professor Noh Junhyuk, the team consisted of integrated master's and doctoral student Lee Jeonghyeon, master's student Kim Hyunseo, and undergraduate intern Jang Hanna. The team presented its research at IEEE FG 2026, held in Kyoto, Japan, at the end of May. Competing against 28 teams from leading universities and research institutions around the world, the Ewha team secured second place, demonstrating its outstanding research capabilities.
The BlEmoRe Challenge is an international competition designed to evaluate how accurately artificial intelligence can recognize blended emotions—complex emotional states that people experience in real-world situations. While conventional emotion recognition research has primarily focused on identifying single emotions such as happiness, sadness, or anger, BlEmoRe addresses the more realistic and challenging task of predicting multiple emotions simultaneously, along with their relative salience.
To support the competition, organizers created a dataset of more than 3,000 video and audio recordings featuring emotional expressions performed by 58 actors. The dataset includes a wide range of emotional expressions encompassing both single and blended emotions.
The research team proposed a novel AI framework that effectively integrates multimodal features, including facial expressions, speech, and behavioral cues. In particular, the team developed a Rank-Aware Selective Fusion method, which selectively identifies and combines the most informative features generated by multiple pre-trained encoders according to the context, significantly improving the accuracy of blended emotion recognition. The framework also introduces a learning architecture that separately models emotion presence and emotional salience before integrating them, enabling more refined analysis of subtle differences among blended emotions.
Photo of integrated master's and doctoral student Lee Jeonghyeon presenting the research.
The study demonstrates that AI can move beyond recognizing a single emotion toward understanding the complex and dynamic emotional states experienced by humans. The findings are expected to have broad applications in AI agents, digital humans, educational and counseling systems, healthcare services, and other fields that require natural human-AI interaction.
"People rarely experience just one emotion at a time. More often, multiple emotions coexist," said Lee Jeonghyeon, who led the research. "This study demonstrates the potential of AI to better understand blended emotions, and we will continue advancing human-centered affective AI technologies."
The research was supported by L4BOX, an AI-based digital content platform company. Working in collaboration with the company, the research team is advancing blended emotion recognition technologies to enable L4BOX's next-generation content services to better understand and respond to users' emotional states. The team plans to further expand this research into interactive content and AI services powered by emotion understanding.
Photo of the competition organizing committee and the winning teams (Lee Jeonghyeon, integrated master's and doctoral student, second from left).