Collaborative multi-instance feature aggregation for visual privacy protection

doi:10.1016/j.patcog.2026.114927

Collaborative multi-instance feature aggregation for visual privacy protection

2026-09-25

0

PRE

AI

DOI：10.1016/j.patcog.2026.114927

Original

Original request for help

Save

Abstract

En 中文

• MIFA resolves privacy-utility trade-offs without injecting artificial noise. • It synthesizes protected data by aggregating multi-instance feature fragments. • Class-specific Shannon entropy guides adaptive uncertainty management. • Maintains high downstream utility while resisting attribute and text-image linkages.

Keywords:

Visual privacy protection

Collaborative feature aggregation

Uncertainty management

Identity obfuscation

Collaborative multi-instance feature aggregation for visual privacy protection

Abstract

Journal

Organization

Cited Papers