System and method for signal reconstruction from incomplete data

1. A method for reconstructing a signal from incomplete data, comprising:

acquiring incomplete signal data;

generating an initial reconstruction of the incomplete signal data;

generating a reconstruction starting from the initial reconstruction by repeating the steps of:

calculating a sparsity transform of the reconstruction;

measuring an approximation of sparsity of the reconstruction by applying an m-estimator to the calculated sparsity transform;

iteratively optimizing the reconstruction to minimize output of the m-estimator and thereby maximizing the approximation of sparsity for the reconstruction; and

providing the optimized reconstruction as a representation of the incomplete data,

wherein each of the above steps is performed by a signal processing device.

2. The method of claim 1 , wherein the sparsity transform is a gradient transform.

3. The method of claim 1 , wherein the sparsity transform is a wavelet transform.

4. The method of claim 1 , wherein the incomplete signal data is acoustic data.

5. The method of claim 1 , wherein the incomplete signal data is image data.

6. The method of claim 1 , wherein the image data is MR (Magnetic Resonance) data.

7. The method of claim 1 , wherein the m-estimator is one or more of a Welsch function, a Cauchy function, a Tukey function, a Huber function, a Fair function, or a Geman function.

8. The method of claim 1 , wherein the approximation of sparsity is measured using a robust error function σ α (x) that is defined as:

σ

α

⁡

(

x

)

=

∑

i

⁢

(

1

-

ⅇ

-

α

⁢

⁢

x

i

2

)

wherein α determines the degree to which sparsity is approximated.

9. The method of claim 8 , wherein minimizing the output of the m-estimator is performed by optimizing:

min

x

⁢

σ

α

⁡

(

A

⁢

⁢

x

)

,

⁢

s

.

t

.

⁢

Bx

=

b

wherein A is the sparsity transform, B is a matrix representing the acquisition basis of the acquired incomplete signal data, b is a vector of measurement for the acquired incomplete signal data, and x is the reconstruction.

10. The method of claim 1 , wherein minimizing the output of the m-estimator includes minimizing the occurrence of outputs of the sparsity transforms that are not very small, wherein what is defined to be very small is determined by α, which is the degree to which sparsity is approximated.

11. The method of claim 1 , wherein minimizing the output of the m-estimator includes minimizing a robust error function σ α (x) that is defined by the m-estimator.

12. The method of claim 1 , wherein minimizing the output of the m-estimator is performed using a series of quadratic minimization problems.

13. The method of claim 1 , wherein the incomplete signal data is assumed to be either noiseless or to include noise and when it is assumed that the incomplete signal data includes noise, noise is penalized during the generation of the initial reconstruction.

14. The method of claim 13 , wherein noise is penalized during the generation of the initial reconstruction includes relaxing a strict equality constraint and penalizing deviations from measurements with an l 2 measure.

15. The method of claim 13 , wherein noise is penalized during the generation of the initial reconstruction by penalizing deviations from measurements with an l 0 measure that measures error as a number of acquired measurements that are not satisfied by the initial reconstruction.

16. The method of claim 1 , wherein the initial reconstruction is generated by assuming data missing from the incomplete data has a particular value.

17. The method of claim 1 , wherein the initial reconstruction is generated using a known approach for finding an optimized reconstruction or by using, as the initial reconstruction, an optimized reconstruction of a prior performance of the steps of calculating the sparsity transform, measuring an approximation of sparsity of the reconstruction, and iteratively optimizing the reconstruction.

18. The method of claim 17 , wherein in the steps of measuring an approximation of sparsity of the reconstruction and iteratively optimizing the reconstruction, sparsity is approximated to a different degree than when these steps are performed during the prior performance.

19. The method of claim 1 , wherein providing the optimized reconstruction as a representation of the incomplete data includes displaying the reconstruction to a user.

20. A signal processing system for reconstructing a signal from incomplete data, comprising:

an initial reconstruction unit for generating an initial reconstruction of incomplete signal data; and

an optimizing unit for generating a reconstruction starting with the initial reconstruction by iteratively optimizing the reconstruction to maximize an approximation of sparsity for the reconstruction,

wherein the approximation of sparsity for the reconstruction includes the use of an m-estimator, and

wherein the initial reconstruction unit and the optimization unit include one or more computer processors.

21. The system of claim 20 , wherein the approximation of sparsity for the reconstruction includes the use of an m-estimator applied to a calculation of a sparsity transform of the reconstruction.

22. A computer system comprising:

a processor; and

a program storage device readable by the computer system, embodying a program of instructions executable by the processor to perform method steps for reconstructing a signal from incomplete data, the method comprising:

generating an initial reconstruction of an incomplete signal data; and

generating a reconstruction starting from the initial reconstruction by repeating the steps of:

calculating a sparsity transform of the reconstruction;

measuring an approximation of sparsity of the reconstruction by applying an m-estimator to the calculated sparsity transform; and

iteratively optimizing the reconstruction to maximize the approximation of sparsity for the reconstruction.