MLR: An Efficient Denoising Model for Highly Corrupted Images

Internet of Things (IoT) consists of devices that generate, process, and exchange vast amounts of images. Unfortunately, these images always contain some kinds of noise, which significantly degrades data utility after the processing center receives the images. Image denoising plays a crucial role to recover the original image approximately from its noisy image according to the features of noise distribution and the structure information of the original image. However, the existing denoising algorithms are invalid when the original images are highly corrupted due to the high computational complexity. Therefore, this paper proposes a novel denoising algorithm based on the multi-low-rank model (MLR), which successively enforces similar blocks, dictionaries, and coefficient matrices approximating to low rank, thereby gradually removing noise. Extensive experimental simulations demonstrate that the MLR algorithm has the optimal denoising performance in terms of denoising quality and efficiency, especially in the case of strong salt noise.