Dynamic patch-level contrastive learning for image dehazing

doi:10.1016/j.patcog.2026.113529

Dynamic patch-level contrastive learning for image dehazing

2026-03-18

0

PRE

AI

DOI：10.1016/j.patcog.2026.113529

Original

Original request for help

Save

Abstract

En 中文

• A patch-based contrastive learning method is proposed for image dehazing. • The method dynamically selects positive and negative samples based on fog density. • Grayscale images are used to reduce atmospheric light interference. • Both local and global features are captured to preserve image details. • The approach achieves strong results on synthetic and real-world datasets.

Keywords:

contrastive learning

image dehazing

patch-based method

fog density

feature preservation

Dynamic patch-level contrastive learning for image dehazing

Abstract

Journal

Organization