Discrete Cosine Stockwell Transform vs. Discrete Cosine Transform in Despeckling OCT images

Hard thresholding through coefficients selection and orthonormal vectors projection is investigated for Discrete Cosine Stockwell transform (DCST) and Discrete Cosine transform (DCT) towards despeckling Optical Coherence Tomography (OCT) images. The whole image is first divided into small dyadic size of sub images, then every sub image is subjected to either DCST or DCT through each of the two mentioned hard thresholding techniques. When applied to simulated and real OCT images, results show the higher efficiency of the projection technique (projecting the noisy signal onto the signal subspace) over the regular coefficients selection technique. Furthermore, projection technique is found more efficient, numerically and visually, in suppressing speckle noise than popular benchmark schemes used in OCT area such as Total Variation (TV) filter, Geometric Non-Linear Diffusion (GNLD) filter and Detail Preserving Anisotropic Diffusion (DPAD) filter.