RAW image reconstruction based on multi-scale attention mechanism
Traditional image signal processing(ISP)algorithms are cumbersome.Therefore,based on the PyNET model that can replace ISP algorithms,an end-to-end RAW image reconstruction method was proposed,called Py-CBAM.This method introduced an efficient attention mechanism and used it to redesign the multi-level and multi-scale structure of the PyNET network to achieve adaptive weighting of features at different scales,so as to improve the image reconstruction performance to a large extent.The experimental results show that the peak signal-to-noise ratio(PSNR)and structural similarity index measure(SSIM)obtained by the proposed method on the publicly available ZRR dataset improve by 0.37 dB and 0.0018 compared with those by the PyNET method.After retraining the Py-CBAM on the ZRR dataset and the newly constructed NRR dataset,the PSNR and SSIM reach 25.73 dB and 0.965 4,respectively.Visually,the proposed method solves the problems of high noise and chromatic aberration and distortion in RAW image reconstruction.It can also enhance the reconstruction ability of the model under different lighting environment conditions in multiple scenes.The reconstruction results are more realistic and have better visual quality,especially in the overexposed and dark areas of the image.
image signal processingimage reconstructionaugmentation networkattention mechanismdeep learning
万方数据
维普
