BCS-SPL—Block Compressed Sensing with Smooth Projected Landweber Reconstruction

Sungkwang Mun and James E. Fowler

>About BCS-SPL

BCS-SPL combines block-based compressed-sensing sampling (BCS) of an image with a smoothed projected-Landweber (SPL) iterative reconstruction. Sampling is driven by random matrices applied on a block-by-block basis, while the reconstruction is a variant of projected Landweber (PL) reconstruction (also known as iterative hard thresholding (IHT)) that incorporates a smoothing operation (Wiener filtering) intended to reducing blocking artifacts. In essence, this filtering operation imposes smoothness in addition to the sparsity inherent to PL.
MC-BCS-SPL is a motion-compensated version of BCS-SPL employing frame-to-frame residual reconstruction for video.
MS-BCS-SPL is a multiscale version of BCS-SPL that deploys BCS within the domain of a wavelet transform.
MH-BCS-SPL is a multihypothesis version of BCS-SPL with reconstruction driven by the measurement-domain residual resulting from multiple predictions culled from neighboring blocks.
BCS-SPL-DPCM applies DPCM and uniform scalar quantization to BCS-SPL to produce a compressed bitstream from the BCS measurements.
MC-CS-PL is a motion-compensated compressed-sensing reconstruction for dynamic MRI (does not use BCS, but adapts PL reconstruction to dynamic MRI).




This material is based upon work supported by the National Science Foundation under Grant No. 0915307. Any opinions, findings and conclusions, or recomendations expressed in this material are those of the author and do not necessarily reflect the views of the National Science Foundation (NSF).

Last update: 22-aug-2021