CAA-Net: End-to-End Two-Branch Feature Attention Network for Single Image Dehazing

In this paper, we propose an end-to-end two-branch feature attention network. The network is mainly used for single image dehazing. consists of two branches, call it CAA-Net: 1) A U-NET composed different-level fusion based on (FEPA) structure and residual dense block (RDB). order to make full use all the hierarchical features image, RDB. RDB contains connected layers local with learning. We also a which called FEPA.FEPA could retain information shallow layer transfer deep layer. FEPA serveral modules (FPA). FPA combines learning channel mechanism pixel mechanism, extract from different channels pixels. 2) several levels structures. weights learn adaptively, give more weight important features. final output result CAA-Net combination branch prediction results. Experimental results show that proposed by us surpasses most advanced algorithms before