reconstruction with adaptive feature-specific imaging
DESCRIPTION
Reconstruction with Adaptive Feature-specific Imaging. Jun Ke 1 and M ark A. Neifeld 1,2. 1 Department of Electrical and Computer Engineering, 2 College of Optical Sciences University of Arizona. Frontiers in Optics 2007. Outline. Motivation for FSI and adaptation. - PowerPoint PPT PresentationTRANSCRIPT
Reconstruction with Adaptive Feature-specific Imaging
Jun Ke1 and Mark A. Neifeld1,2
1Department of Electrical and Computer Engineering,
2College of Optical Sciences
University of Arizona
Frontiers in Optics 2007
Outline
Frontiers in Optics 2007
Motivation for FSI and adaptation.
Adaptive FSI using PCA/Hadamard features.
Adaptive FSI in noise.
Conclusion.
Motivation - FSI Reconstruction with Feature-specific Imaging (FSI) :
Frontiers in Optics 2007
FSI benefits:
Lower hardware complexity
Smaller equipment size/weight
Higher measurement SNR
High data acquisition rate
Lower operation bandwidth
Less power consumption
1myn mM 1ˆ nx1nxnmF 1my
Reconstruction matrix M (nxm)
x̂
object
object estimate
)1( nx
DMD
Imaging optics Imaging
optics
single detector
)1( nif
feature T
iiy fx
),,2,1( Mi
Sequential architecture:
Parallel architecture:
LCD
LCD
LCD
1f
2f
Mf
)1( nx
1T
1 fxy
2T
2 fxy
MMy fxT
M (nxm)
Motivation - Adaptation
Frontiers in Optics 2007
Acquire feature measurements sequentially
Use acquired feature measurements and training data to adapt the next projection vector
The design of projection vector effects reconstruction quality.
Estimation
Poorly designed projection vectors
Testing sample
Training samples
Projection axis 2
Static PCA
Projection axis 1
Using PCA projection as example Well designed projection vectorsAdaptive PCA
Estimation
Projection axis 2
Projection value
Training samples
for 2nd projection vector
Projection axis 1
Frontiers in Optics 2007
1...
ˆ i m mm i
y
x fObject estimate
yi = fiTx
Calculate fi+1
Reconstruction
Object x
Update Ai to Ai+1
according to yi
Computational Optics
Calculate f1
Ri+1
Calculate R1 from A1
Adaptive FSI (AFSI) – PCA:
i: adaptive step index
Ai: ith training set
Ri: autocorrelation matrix of Ai
fi: dominate eigenvector of Ai
yi: feature value measured by fi High diversity of training data helps adaptation
PCA-Based AFSI
Testing sample
K(1) nearest samplesProjection axis
Testing sample
K(1) nearest samples
Selected samples
According to 1st feature
According to 2nd feature
K (2) nearest samples
Projection axis 2
Projection axis 1
Object examples (32x32):
Tx̂ = F y Reconstructed object:
2ˆ{|| || }/E N x - x RMSE:
y = Fx Feature measurements: where, : 1 : N M N x F
is the total # of featuresM
PCA-Based AFSI
Frontiers in Optics 2007
Number of training objects: 100,000
Number of testing objects: 60
RMSE reduces using more features
RMSE reduces using AFSI compare to static FSI
Improvement is larger for high diversity data
RMSE improvement is 33% and 16% for high and low diversity training data, when M = 250.
Frontiers in Optics 2007
AFSI – PCA:
PCA-Based AFSI
Reconstruction from static FSI (M = 100)
Reconstruction from AFSI (M = 100)
K increases
Projection vector’s implementation order is adapted.
Frontiers in Optics 2007
xi(mean): average vector of Ai
fi: dominant Hadamard vector for Ai
AFSI – Hadamard:
Hadamard-Based AFSI
Object estimate
yiL+j = f iL+jTx
(j=1,…,L)
Choose fiL+1 ~ f(i+1)L
ReconstructionObject x
Update Ai to Ai+1
according to yiL+j
Computational Optics
Choose f1~fL
xi+1(mean)
x1(mean)
Sort
SortHadamard
bases
Selected samples
K(1) nearest samples
testing sample
projection axis 1
K(1) nearest samples
testing sample
projection axis 2
K(2) nearest samples
sample mean
First 5 Hadamard basis←Static FSI AFSI→
according to 1st feature
according to 2nd feature
sample mean
projection axis 1
RMSE reduces in AFSI compared with static FSI
RMSE improvement is 32% and 18% for high and low diversity training data, when M = 500 and L = 10.
AFSI has smaller RMSE using small L when M is also small
AFSI has smaller RMSE using large L when M is also large
Hadamard-Based AFSI
Frontiers in Optics 2007
AFSI – Hadamard:
Reconstruction from adaptive FSI
Reconstruction from static FSI
K increases
L d
ec
rea
se
s
L in
cre
as
es
Hadamard-Based AFSI – Noise
Frontiers in Optics 2007
AFSI – Hadamard:
Hadmard projection is used because of its good reconstruction performance
Feature measurements are de-noised before used in adaptation
Auto-correlation matrix is updated in each adaptation step
Wiener operator is used for object reconstruction
nFxnyy 0
1)( nyyy RRRM
{ }
{( )( ) }
Ty
T
Tx
E
E
R yy
Fx Fx
FR F
1...ˆ i m mm i
y
x fObject estimate
yiL+j = fiL+jTx+niL+j
(j = 1,2,…L)
Choose fiL+1~f(i+1)L
Reconstruction
Object x
Update Ai to Ai+1
according to
Computational OpticsChoose f1~fL
xi+1(mean)
Calculate x1(mean)
from de-noising yiL+j
Calculate Ri for Ai
T : integration time
σ0
2 = 1
ˆiL jy
Sort Hadamard bases
Sort
ˆiL jy
Frontiers in Optics 2007
RMSE in AFSI is smaller than in static FSI
RMSE is reduced further by modifying Rx in each adaptation step
RMSE improvement is larger using small L when M is also small
RMSE is small using large L when M is also large
Hadamard-Based AFSI – Noise
High diversity training data; σ02 = 1
T : integration time/per feature
σ0
2 = 1
detector noise variance σ2 = σ02 /T
High diversity training data; σ02 = 1
K increases
L d
ec
rea
se
s
L in
cre
as
es
T : integration time/per feature; M0: the number of features
Total feature collection time = T × M0
RMSE reduces as T increases
High reconstruction quality requirement needs longer total feature collection time
To achieve each RMSE requirement, there is a minimum total feature collection time.
Hadamard-Based AFSI – Noise
Frontiers in Optics 2007
High diversity training data; σ02 = 1High diversity training data; σ0
2 = 1
Conclusion
Frontiers in Optics 2007
Noise free measurements:
PCA-based and Hadmard-based AFSI system are presented
AFSI system presents lower RMSE than static FSI system
Noisy measurements:
Hadamard-based AFSI system in noise is presented
AFSI system presents smaller RMSE than static FSI system
There is a minimum total feature collection time to achieve a reconstruction quality requirement