Low Tensor Train- and Low Multilinear Rank Approximations for De-speckling and Compression of 3D Optical Coherence Tomography Images
This paper proposes low tensor-train (TT) rank and low multilinear (ML) rank approximations for de-speckling and compression of 3D optical coherence tomography (OCT) images for a given compression ratio (CR). To this end, we derive the alternating direction method of multipliers based algorithms for the related problems constrained with the low TT- and low ML rank. Rank constraints are implemented through the Schatten-p (Sp) norm, p e {0, 1/2, 2/3, 1}, of unfolded matrices. We provide the proofs of global convergence towards a stationary point for both algorithms. Rank adjusted 3D OCT image tensors are finally approximated through tensor train- and Tucker alternating least squares decompositions. We comparatively validate the low TT- and low ML rank methods on twenty-two 3D OCT images with the JPEG2000 and 3D SPIHT compression methods, as well as with no compression 2D bilateral filtering (BF), 2D median filtering (MF), and enhanced low-rank plus sparse matrix decomposition (ELRpSD) methods. For the CR<10, the low Sp TT rank method with pe{0, 1/2, 2/3} yields either highest or comparable signal-to-noise ratio (SNR), and comparable or better contrast-to-noise ratio (CNR), mean segmentation errors (SEs) of retina layers and expert-based image quality score (EIQS) than original image and image compression methods. It compares favorably in terms of CNR, fairly in terms of SE and EIQS with the no image compression methods. Thus, for CR<10 the low S2/3 TT rank approximation can be considered a good choice for visual inspection based diagnostics. For 2<CR<60, the low S1 ML rank method compares favorably in terms of SE with image compression methods and with 2D BF and ELRpSD. It is slightly inferior to 2D MF. Thus, for 2<CR<60, the low S1 ML rank approximation can be considered a good choice for segmentation based diagnostics either on-site or in the remote mode of operation.