Image denoising using min-max filtering and self-attention neural networks
Yen-Yu Lu, Hung-Chin Jang · IET conference proceedings. · 2025
Digital images may be subjected to electromagnetic interference during transmission and camera component damage, resulting in image quality degradation due to pulse noise interference. How to effectively restore the images interfered with by salt-and-pepper (SAP) noise is essential to improve the quality of digital images. This paper proposes using minimum-maximum filtering and self-attention neural networks (SANN) to remove the SAP noise in the deteriorated images. Firstly, the noise density of the image is estimated. If it is a slight or mildly noisy image, the noise pixels are reconstructed using a SANN. Conversely, if it is a medium or high noise-density interference, max-pooling and min-pooling, through several iterations, generate min-max pooling. Finally, min-max pooled images are composed by mean filtering. The experimental results demonstrate that the proposed method can effectively reconstruct the image interfered with by various noise densities and outperform many state-of-the-art algorithms by visualization and objective measurement tests.