Scholarlink

This study aims to enhance the efficacy of the VR-based Self-Attachment Technique (SAT) through the implementation of a Facial Expression Recognition (FER) algorithm. SAT is an innovative self-administered psychotherapeutic technique wherein individuals engage with their childhood self, represented by their avatars, to improve their capacity for emotion self-regulation. Accurate emotion recognition is vital for the customisation of the procedure, but occlusion caused by the VR headset poses a significant challenge. Drawing inspiration from the EmoFAN architecture, we propose a novel model tailored for VR-occluded emotion recognition. Our model demonstrated improvements and adaptability in valence-arousal estimation and discrete emotion classification, particularly on datasets such as AffectNet and AFEW-VA, wherein image data underwent automated occlusion. Comparative analysis against other models (EmoFAN-VR and EmoFAN) tested on occluded images showcased enhanced performance across various metrics.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Facial Expression Recognition on VR-Occluded Images for Personalised Self-Attachment Intervention

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Neophytos Polydorou,
  • Abbas Edalat

摘要

This study aims to enhance the efficacy of the VR-based Self-Attachment Technique (SAT) through the implementation of a Facial Expression Recognition (FER) algorithm. SAT is an innovative self-administered psychotherapeutic technique wherein individuals engage with their childhood self, represented by their avatars, to improve their capacity for emotion self-regulation. Accurate emotion recognition is vital for the customisation of the procedure, but occlusion caused by the VR headset poses a significant challenge. Drawing inspiration from the EmoFAN architecture, we propose a novel model tailored for VR-occluded emotion recognition. Our model demonstrated improvements and adaptability in valence-arousal estimation and discrete emotion classification, particularly on datasets such as AffectNet and AFEW-VA, wherein image data underwent automated occlusion. Comparative analysis against other models (EmoFAN-VR and EmoFAN) tested on occluded images showcased enhanced performance across various metrics.