Optical Coherence Tomography Noise Reduction Over Learned Dictionaries with Introduction of Complex Wavelet for start dictionary

摘要

Optical Coherence Tomography (OCT) suffers from speckle noise which causes erroneous interpretation. OCT denoising methods may be studied in "raw image domain" and "sparse representation". Comparison of mentioned denoising strategies in magnitude domain shows that wavelet-thresholding methods had the highest ability, among which wavelets with shift invariant property yielded better results. We chose dictionary learning to improve the performance of available wavelet-thresholding by tailoring adjusted dictionaries instead of using pre-defined bases. Furthermore, in order to take advantage of shift invariant wavelets, we introduce a new scheme in dictionary learning which starts from a dual tree complex wavelet. we investigate the performance of different speckle reduction methods: 2D Conventional Dictionary Learning (2D CDL), Real part of 2D Dictionary Learning with start dictionary of dual-tree Complex Wavelet Transform (2D RCWDL) and Imaginary part of 2D Dictionary Learning with start dictionary of dual-tree Complex Wavelet Transform (2D ICWDL). It can be seen that the performance of the proposed method in 2D R/I CWDL are considerably better than other methods in CNR.