Abstract:In low-light environments, the signal-to-noise ratio (SNR) of complementary metal-oxide-semiconductor (CMOS) image sensors decreases significantly. In order to improve the imaging quality, this paper proposes a transformer-based denoising method, illumination prompting dual-attention transformer (IPDT). In the noise model, the effect of dark current noise on the low light images is emphasized. The denoising network contains the dual-attention transformer block (DTB) module including spatial-channel attention and the prompt-block (PB). The DTB captures global and detail information of low-light images through cross-window and cross-channel interactions, respectively. The PB acquires illumination information through learnable prompt parameters, dynamically guiding the denoising network. Experiments on see-in-the-dark (SID) dataset and smartphone image denoising dataset (SIDD) indicate that the proposed method effectively suppresses the noise of low-light images and demonstrates superior performance in terms of blind referenceless image spatial quality evaluator (BRISQUE) and multi-scale structural similarity index (MS-SSIM) metrics.