Abstract:Purpose Auto-calibration parallel imaging (acPI) [1] is based on local correlations in k-space. It is known to perform robustly in practice, especially when accurate sensitivity information is hard to obtain. Also, acPI renders new opportunities for motion correction [2]. However, the reconstruction quality of acPI methods is highly dependent on the accuracy of the interpolation kernels calculated from the auto calibration signal (ACS). Corruption of ACS data, e.g. by motion, often leads to serious artifacts in the reconstructed images. In this work, we propose to exploit the redundancy in k-space to detect and correct sparse corruptions in ACS data, which could result from random, time-limited motion in clinical practice (e.g. swallowing, jerk, etc). Our work is based on low-rank matrix completion with sparse errors, and is an extension of our calibrationless parallel imaging reconstruction methods [3,4] . Methods Low-rank matrix completion: In general, missing entries of a matrix can be completed if the matrix is low-rank. This could be efficiently achieved by singular-value thresholding (SVT) [5]. In acPI, the calibration matrix should have low-rank [3,4]. However, if corrupted by random and sparse errors, the correlation within the calibration matrix will be reduced, leading to higher rank values. Intuitively, we could enforce low-rank regularization onto the corrupted calibration matrix to detect and correct these errors. For consistency with the uncorrupted acquired data, we use a soft-thresholding scheme to control the difference between original k-space and newly synthesized k-space by low-rank approximation. The flowchart of our proposed method is shown in Figure 1 and can be described as: (1) Construct matrix A from over-lapping blocks; (2) Compute [ , , ] ( ) U V svd A Σ = ; Threshold the singular values ( , ) S λ Σ = Σ ; (3) Compute ' A U V = Σ ;

Exploiting temporal stability and low-rank structure for motion capture data refinement

A 3D Human Motion Refinement Method Based on Sparse Motion Bases Selection.

Robust Visual Tracking Via CAMShift and Structural Local Sparse Appearance Model

Predicting Missing Markers In Human Motion Capture Using L1-Sparse Representation

Mining Spatial-Temporal Patterns and Structural Sparsity for Human Motion Data Denoising

Estimating small structural motions based on sparsity enforcement

Dynamic Matrix Recovery

Motion Complement and Temporal Multifocusing for Skeleton-Based Action Recognition

Cross Refinement Techniques for Markerless Human Motion Capture

A target-free video structural motion estimation method based on multi-path optimization

Towards Robust and Accurate Single-View Fast Human Motion Capture

Sparse moving factorization for subspace video stabilization

Generalized Low-Rank Matrix Completion Model with Overlapping Group Error Representation

Robust Low-rank Matrix Completion for sparse motion correction in auto calibration PI

Online high rank matrix completion

CP decomposition-based algorithms for completion problem of motion capture data

Enhanced Spatio-Temporal Context for Temporally Consistent Robust 3D Human Motion Recovery from Monocular Videos

Non-rigid Structure-from-Motion: Temporally-smooth Procrustean Alignment and Spatially-variant Deformation Modeling

Human Motion Estimation from a Reduced Marker Set.

Spatio-Temporal Online Matrix Factorization for Multi-Scale Moving Objects Detection

Dynamic Multi-Person Mesh Recovery From Uncalibrated Multi-View Cameras