Adaptive Weighted Nuclear Norm Minimization for Removing Speckle Noise from Optical Coherence Tomography Images

Low-rank matrix approximation is widely used in various fields of computer science, and weighted nuclear norm minimization (WNNM) has demonstrated improved results by shrinking the different weights of singular values. In this paper, an adaptive WNNM is proposed, considering the relative significance of image information by modifying the WNNM. As a result, singular values that contain more important information are relatively saved, whereas those that contain less crucial information are drastically reduced. When applying this method to noised image with black and white dot-noised images, the algorithm showed improved performance in both instances. Especially, when applied to images with white dot noise, the denoised results were outstanding. In addition, the proposed algorithm was successfully applied onto the optical coherence tomography images, numerically and visually.