Double-Branch Dehazing Network based on Self-Calibrated Attentional Convolution

Abstract

Single image dehazing is an essential preprocessing task to improve the image’s visual effect. In this paper, we propose a Double-Branch Dehazing Network based on Self-Calibrated Attentional Convolution (DBDN-SCAC), which effectively captures the multi-scale features of important information for solving dehazing problems. More specifically, our proposed DBDN-SCAC uses an encoder–decoder architecture based on a Laplacian Pyramid to learn information at different scales. At the same time, the self-calibration attention convolution module is incorporated after each scaling to expand the receptive field and enrich the output feature. Furthermore, to avoid information loss caused by downsampling, a new self-calibration branch is established to help the image recover more detailed information. Finally, a Content-aware Upsampling module is added in two branches to allow the network in reconstructing better images. The effectiveness of our approach, which is compared with the state-of-the-art methods in the public datasets, is justified by our achieving the best results.