Speckle noise reduction for optical coherence tomography images via non-local weighted group low-rank representation

In this work, a non-local weighted group low-rank representation (WGLRR) model is proposed for speckle noise reduction in optical coherence tomography (OCT) images. It is based on the observation that the similarity between patches within the noise-free OCT image leads to a high correlation between them, which means that the data matrix grouped by these similar patches is low-rank. Thus, the low-rank representation (LRR) is used to recover the noise-free group data matrix. In order to maintain the fidelity of the recovered image, the corrupted probability of each pixel is integrated into the LRR model as a weight to regularize the error term. Considering that each single patch might belong to several groups, and multiple estimates of this patch can be obtained, different estimates of each patch is aggregated to obtain its denoised result. The aggregating weights are exploited depending on the rank of each group data matrix, which can assign higher weights to those better estimates. Both qualitative and quantitative experimental results on real OCT images show the superior performance of the WGLRR model compared with other state-of-the-art speckle removal techniques.