Multiscale Residual and Attention Guidance for Low-Light Image Enhancement in Visual SLAM

Low-light image enhancement is one of the challenges in robotics. For low-light images, only adjusting the brightness of the image will inevitably amplify the noise and cause more severe color distortion, which is contrary to the original purpose. To balance multiple factors, including brightness, color, and noise, we propose a simple and efficient lighting network of multiscale residual and attention guidance (MRAG) to enhance low-light images. The MRAG network model is founded on Retinex theory and includes a decomposition layer that separates low-light images into reflectance and illumination maps, as well as a reflectance recovery layer and an illumination recovery layer. In the decomposition layer, an end-to-end attention mechanism based on multibranch convolutional neural networks is adopted, and in the reflectance layer, a multiscale residual network is used to fully learn image features. Extensive experiments on the LIME and NPE data sets show that the proposed method outperforms state-of-the-art methods in terms of image quality evaluator, with respective improvements of 3.9% and 8.9%. The experiments on the LOL data set demonstrate that the proposed method has enhancements in the peak signal-to-noise ratio, structural similarity index measure, lightness order error, and natural image quality evaluator evaluation metrics compared to state of the art. Compared with state-of-the-art low-light enhancement methods, the proposed method has some improvements in optimizing the SLAM front end.