Download presentation
Presentation is loading. Please wait.
1
An Introduction to Compressive Sensing
Speaker: Ying-Jou Chen Advisor: Jian-Jiun Ding
2
Compressive Compressed
Sensing Sampling CS
3
Outline Conventional Sampling & Compression Compressive Sensing
Why it is useful? Framework When and how to use Recovery Simple demo
4
Sampling and Compression
Reviewโฆ Sampling and Compression
5
Nyquistโs Rate Perfect recovery ๐ ๐ โฅ2 ๐ ๐
6
Transform Coding Assume: signal is sparse in some domainโฆ
e.g. JPEG, JPEG2000, MPEGโฆ Sample with frequency ๐ ๐ . Get signal of length N Transform signal ๏ K (<< N) nonzero coefficients Preserve K coefficients and their locations ็ซๅ่ฌไธไธ
7
Compressive Sensing
8
Compressive Sensing Sample with rate lower than ๐ ๐ !!
Can be recovered PERFECTLY! ๅฎๅฎwith lower rate ไธๅฒๅฎณ ่ฝๅค ้ๅๆๆฏ็็ๅพๅฑ
9
Comparison Nyquistโs Sampling Compressive Sensing Sampling Frequency
โฅ 2๐ ๐ < 2๐ ๐ Recovery Low pass filter Convex Optimization
10
Some Applications ECG One-pixel Camera Medical Imaging: MRI
11
ฮฆ Framework ๐ฒ= ๐ฝ๐ = ๐ฆ ๐ N M N M
N: length for signal sampled with Nyquistโs rate M: length for signal with lower rate ฮฆ: Sampling matrix
12
When? How? Two things you must knowโฆ
13
Whenโฆ. Signal is compressible, sparseโฆ ๐ฆ ๐ N ๐ฅ ฮฆ = M M N ฮจ
14
Exampleโฆ ECG ๐: ๅฟ้ปๅ่จ่ ฮจ: DCT (discrete cosine transform)
15
ฮฆ ฮจ Howโฆ How to design the sampling matrix?
How to decide the sampling rate (M)? ๐ฆ N ๐ฅ ฮฆ = M ฮจ
16
Sampling Matrix Low coherence Low coherence ๐ฆ ๐ฅ ฮฆ = ฮจ
17
Coherence Describe similarity ๐ ๐ฝ,๐ฟ =๐งโ ๐ฆ๐๐ฑ ๐ ๐ค , ๐ ๐ฃ ๐
๐ ๐ฝ,๐ฟ =๐งโ ๐ฆ๐๐ฑ ๐ ๐ค , ๐ ๐ฃ ๐ High coherence ๏ more similar Low coherence ๏ more different ๐ ๐ฝ,๐ โ[1,๐]
18
Example: Time and Frequency
For example, ๐บ๐๐๐๐ ๐๐๐๐๐ and ๐ญ๐๐๐๐๐๐ ๐๐๐๐๐ ๐ ๐ =๐ฟ(๐กโ๐), โ ๐ = 1 ๐ ๐ ๐ 2๐ ๐๐ก/๐
19
Fortunatelyโฆ Random Sampling Low coherence with deterministic basis.
iid Gaussian N(0,1) Random ยฑ1 Low coherence with deterministic basis.
20
More about low coherenceโฆ
Random Sampling
21
Sampling Rate Theorem ๐ฆโฅ๐โ ๐ ๐ ๐ฝ,๐ฟ โ๐โ ๐ฅ๐จ๐ ๐ง
Can be exactly recovered with high probability. Theorem ๐ฆโฅ๐โ ๐ ๐ ๐ฝ,๐ฟ โ๐โ ๐ฅ๐จ๐ ๐ง C : constant ฮผ: coherence S: sparsity n: signal length
22
BUTโฆ. ฮฆ ฮจ Recovery y= ฮฆf ๐๐จ๐ฅ๐ฏ๐ ๐๐จ๐ซ x f= ฮฆ โ1 y s.t. y= ฮฆฮจx = ๐ฆ ๐ N ๐ฅ M
23
โ 1 Recovery Many related researchโฆ GPSR
(Gradient projection for sparse reconstruction) L1-magic SparseLab BOA (Bound optimization approach) โฆ..
24
Total Procedure ๅทฒ็ฅ: ๐ฒ , ๐ฝ Sampling (Assume f is spare somewhere)
Find an incoherent matrix ฮฆ e.g. random matrix f Sample signal y=ฮฆf ๅทฒ็ฅ: ๐ฒ , ๐ฝ ๐๐๐ ๐ ๐๐๐ ๐ ๐ ๐ .๐ก. ๐ฒ=๐ฏ ๐ฌ ๐ฑ =๐ ๐ฌ Recovering
25
Sum up ๆ size ็บ nx1 ๅจๆdomain ไธ sparse็่จ่
็จsize็บ mxn ็ random matrix ๅsampling (m<n) ๅพๅฐ size ็บ mx1 ็measurement y ๅฐ y ๅ L1 norm recovery ้ๅๅพๅฐ x_recovery
26
Demo Time
27
Reference Candes, E. J. and M. B. Wakin (2008). "An Introduction To Compressive Sampling." Signal Processing Magazine, IEEE 25(2): Baraniuk, R. (2008). Compressive sensing. Information Sciences and Systems, CISS nd Annual Conference on. Richard Baraniuk, Mark Davenport, Marco Duarte, Chinmay Hegde. An Introduction to Compressive Sensing.
28
Thanks a lot!
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.