Attention-guided Low light enhancement CNN Network

抄録

Low illumination image enhancement is a difficult but scientific task. With the image brightness increasing, the noises are amplified, and with the contrast and detail increasing, the false information is generated. To solve this problem, a multi-branch attention network is proposed to process low-light images directly without additional operations. The proposed network is composed with enhancement module (EM) and Convolutional Block Attention Module (CBAM). The attention module can make the CNN network structure gradually focus on the weak light area in the image, and the enhancement module can fully highlight the multi-branch feature graph under the guidance of attention. In this way, the overall quality of the picture will be greatly improved, including contrast, brightness, etc. Through a large number of experiments, our model can produce better visual effects, and also achieve good results in quantitative indicators.