Self Supervised Image Denoising

Self-supervised image denoising aims to restore images from noisy observations without relying on paired clean-noisy image data, addressing the significant cost and difficulty of obtaining such datasets. Current research focuses on improving model architectures, such as blind-spot networks (BSNs) and transformers, to effectively prevent trivial solutions and enhance denoising performance, often incorporating techniques like multi-masking, adaptive supervision, and knowledge distillation. These advancements are crucial for practical applications where clean image data is scarce, particularly in real-world scenarios like raw image processing and biomedical microscopy, enabling more robust and efficient image restoration.