Speckle reduction in optical coherence tomography images of human finger skin by wavelet modified BM3D filter

Speckle pattern, which is inherent in coherence imaging, degrades the visual quality of optical coherence tomography (OCT) images. Speckle noise reduction is an important post-processing step for OCT imaging. As multiplicative noise pattern, speckle noise is difficult to suppress efficiently. Current well know speckle removal techniques are not able to achieve the goals both on speckle suppression and on edge details preservation. To address this issue, a novel speckle noise reduction algorithm is proposed. The algorithm is based on block-matching 3D filter modified by morlet wavelet decomposition. Original OCT image data transformed by logarithmic compression is decomposed into 10 components by morlet wavelet for three levels. Each component is proposed by a suited BM3D filter and the output image is reconstructed by wavelet reverse transformation. Experiment to in vivo human index finger skin OCT images showed that the proposed algorithm achieves good performances in terms of signal-to-noise ratio, equivalent number of looks, contrast-to-noise ratio, edge preservation coefficient, and CPU time compared to other recent high performance image speckle denoising algorithms. Visual comparisons also show that the proposed algorithm provides effective speckle noise suppression while preserving edge sharpness and improving morphological details visibility, such as different layers in finger skin OCT images.