26
Recovering low rank and sparse matrices from compressive measurements Aswin C Sankaranarayanan Rice University Richard G. Baraniuk Andrew E. Waters

Recovering low rank and sparse matrices from compressive measurements

  • Upload
    nancy

  • View
    60

  • Download
    0

Embed Size (px)

DESCRIPTION

Recovering low rank and sparse matrices from compressive measurements. Aswin C Sankaranarayanan Rice University. Richard G. Baraniuk. Andrew E. Waters. Background subtraction in surveillance videos. s tatic camera with foreground objects. r ank 1 background. s parse foreground. - PowerPoint PPT Presentation

Citation preview

Page 1: Recovering low rank and sparse matrices from compressive measurements

Recovering low rank and sparse matrices from compressive measurements

Aswin C SankaranarayananRice University

Richard G. Baraniuk Andrew E. Waters

Page 2: Recovering low rank and sparse matrices from compressive measurements

Background subtraction in surveillance videos

static camera with foreground objects

rank 1 background

sparse foreground

Page 3: Recovering low rank and sparse matrices from compressive measurements

More complex scenarios

Changing illumination + foreground motion

Page 4: Recovering low rank and sparse matrices from compressive measurements

More complex scenarios

Changing illumination + foreground motion

Set of all images of a convex Lambertian scene under changing illumination is very close to a 9-dimensional subspace

[Basri and Jacobs, 2003]

Page 5: Recovering low rank and sparse matrices from compressive measurements

More complex scenarios

Changing illumination + foreground motion

Video can be represented as a sum of a rank-9 matrix and a sparse matrix

Can we use such low rank+sparse model in a compressive recovery framework ?

Page 6: Recovering low rank and sparse matrices from compressive measurements

Hyperspectral cube

450nm 550nm 720nm490nm 580nm

Rank approximately equal number of materials in the sceneData courtesy Ayan Chakrabarti, http://vision.seas.harvard.edu/hyperspec/

Page 7: Recovering low rank and sparse matrices from compressive measurements

Robust matrix completion

low rank matrix

low rank matrix with missing entries

Page 8: Recovering low rank and sparse matrices from compressive measurements

Robust matrix completion

missing + corrupted

entries

low rank matrix

sparse corruptions

Page 9: Recovering low rank and sparse matrices from compressive measurements

Problem formulation• Noisy compressive measurements

L: r-rank matrix S: k-sparse matrix

• Measurement operator is different for different problems– Video CS: operates on each column of the matrix individually– Matrix completion: sampling operator– Hyperspectral

Page 10: Recovering low rank and sparse matrices from compressive measurements

Problem formulation• Noisy compressive measurements

L: r-rank matrix S: k-sparse matrix

Page 11: Recovering low rank and sparse matrices from compressive measurements

Side note: Robust PCA “?”

• Recovery a low rank matrix L and a sparse matrix S, given M = L + S

Robust PCA [Candes et al, 2009] Rank-sparsity incoherence [Chandrasekaran et al, 2011]

• We are interested in recovering a low rank matrix L and a sparse matrix S --- not from M --- but from compressive measurements of M

Page 12: Recovering low rank and sparse matrices from compressive measurements

Connections to CS and Matrix Completion

• If we “remove” L from the optimization, then this reduces to traditional compressive recovery problem

• Similarly, if we “remove” S, then this reduces to the Affine rank minimization problem

Page 13: Recovering low rank and sparse matrices from compressive measurements

Problem formulation

• Key questions– When can we recover L and S ?– Measurement bounds ?– Fast algorithms ?

Page 14: Recovering low rank and sparse matrices from compressive measurements

SpaRCS• SpaRCS: Sparse and low Rank recovery from CS

– A greedy algorithm– It is an extension of CoSaMP [Tropp and Needell, 2009] and ADMiRA [Lee and Bresler, 2010]

Page 15: Recovering low rank and sparse matrices from compressive measurements

SpaRCS• SpaRCS: Sparse and low Rank recovery from CS

– A greedy algorithm– It is an extension of CoSAMP [Tropp and Needell, 2009] and ADMiRA [Lee and Bresler, 2010]

Page 16: Recovering low rank and sparse matrices from compressive measurements

SpaRCS• SpaRCS: Sparse and low Rank recovery from CS

– A greedy algorithm– It is an extension of CoSaMP [Tropp and Needell, 2009] and ADMiRA [Lee and Bresler, 2010]

• Claim– If satisfies both RIP and rank-RIP with small

constants,– and the low rank matrix is sufficiently dense, and sparse

matrix has random support (or bounded col/row degree)– then, SpaRCS converges exponentially to the right

answer

Page 17: Recovering low rank and sparse matrices from compressive measurements

Phase transitions

• p = number of measurements• r = rank, K = sparsity• Matrix of size N x N; N = 512

r=5 r=10 r=15 r=20 r=25

Page 18: Recovering low rank and sparse matrices from compressive measurements

Performance

Run time

CS IT: An alternating projection algorithm that uses soft thresholding at each step

CS APG: Variant of APG for RobustPCA problem.

Accuracy

Page 19: Recovering low rank and sparse matrices from compressive measurements

Video CS

(a) Ground truth

(b) Estimated low rank matrix

(c) Estimated sparse component

Video: 128x128x201Compression 6.67xSNR = 31.1637 dB

Page 20: Recovering low rank and sparse matrices from compressive measurements

Video CS

(a) Ground truth

(b) Compression 3x

Video 64x64x239Compression 3xSNR = 23.9 dB

Page 21: Recovering low rank and sparse matrices from compressive measurements

Hyperspectral recovery results

128x128x128 HS cubeCompression 6.67xSNR = 31.1637 dB

Page 22: Recovering low rank and sparse matrices from compressive measurements

Matrix completion

Run time

CVX: Interior point solver of convex formulation

OptSpace: Non-robust MC solver

Accuracy

Page 23: Recovering low rank and sparse matrices from compressive measurements

Open questions

• Convergence results for the greedy algorithm

• Low rank component is sparse/compressible in a wavelet basis– Is it even possible ?

Page 24: Recovering low rank and sparse matrices from compressive measurements

CS-LDS• [S, et al., SIAM J. IS*]

• Low rank model– Sparse rows (in a wavelet

transformation)

• Hyper-spectral data– 2300 Spectral bands– Spatial resolution 128 x 64– Rank 5

Ground Truth

2% 1%

25.2 dB 24.7 dB

Page 25: Recovering low rank and sparse matrices from compressive measurements

Ground truth512x256x360 voxels

M/N = 10% M/N = 2% M/N = 1%(rank = 20)

Page 26: Recovering low rank and sparse matrices from compressive measurements

Open questions• Convergence results for the greedy algorithm

• Low rank matrix is sparse/compressible in a wavelet basis– Is it even possible ?

• Streaming recovery etc…

dsp.rice.edu