Infrared Image Denoising Based on Dynamic Pyramid and Attention Mechanism

Infrared images are widely utilized across various applications; however, sensor-induced background noise often results in a low signal-to-noise ratio and poor visual quality. To address issues such as incomplete noise removal, high computational cost, and loss of texture features in existing denoising algorithms, this study proposes an infrared image denoising method based on a dynamic pyramid structure and attention mechanisms. First, multi-scale image features are extracted using a pyramid architecture. Second, a dynamic correction and fusion mechanism is introduced to enhance the network's capability for multi-scale feature integration. Finally, a local-context attention block is designed to enhance the restoration of both local details and contextual information. Experimental results on both visible-light and infrared image datasets demonstrate that the proposed algorithm effectively removes noise, preserves texture details, avoids artifacts and speckle noise, and reduces GFLOPS by 65% compared with the NAFNet method.