Scholarlink

<p>In mental health care, fear of social stigma can prevent people from seeking therapy. Video-based remote therapy can lower this barrier, but for highly stigmatized or legally sensitive conditions, additional privacy protection may be required. Our prior work used a landmark “mask” representation for emotion-preserving facial anonymization; in this extension, we introduce humanoid avatarization and directly compare both approaches. In an online user study (N = 33), participants performed open-set human re-identification and human emotion recognition on anonymized videos and reported preferences in an A/B comparison. Participants rated the avatar as more natural and expressive, while rating the mask as more anonymous and private. In contrast, objective re-identification accuracy was lower for the avatar than for the mask (13.46% vs. 20.20%), indicating stronger privacy protection by avatarization despite lower perceived anonymity. Emotion recognition was generally higher with the avatar method, consistent with participants’ utility-oriented preferences.</p>

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

Avatar-Based Anonymization in Remote Therapy: Balancing Privacy and Emotional Expressivity

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月20日

查看全文
  • Jan Marquenie,
  • Carlos Franzreb,
  • Robert Brunstein,
  • Ingo Siegert

摘要

In mental health care, fear of social stigma can prevent people from seeking therapy. Video-based remote therapy can lower this barrier, but for highly stigmatized or legally sensitive conditions, additional privacy protection may be required. Our prior work used a landmark “mask” representation for emotion-preserving facial anonymization; in this extension, we introduce humanoid avatarization and directly compare both approaches. In an online user study (N = 33), participants performed open-set human re-identification and human emotion recognition on anonymized videos and reported preferences in an A/B comparison. Participants rated the avatar as more natural and expressive, while rating the mask as more anonymous and private. In contrast, objective re-identification accuracy was lower for the avatar than for the mask (13.46% vs. 20.20%), indicating stronger privacy protection by avatarization despite lower perceived anonymity. Emotion recognition was generally higher with the avatar method, consistent with participants’ utility-oriented preferences.