Method and system for coding signals using distributed coding and non-monotonic quantization

Patent metadata

Field	Value
Publication number	US-9501717-B1
Application number	US-201514822212-A
Country	US
Kind code	B1
Filing date	Aug 10, 2015
Priority date	Aug 10, 2015
Publication date	Nov 22, 2016
Grant date	Nov 22, 2016

Abstract

A method reconstructs and an uncompressed signal by first obtaining an encoded signal corresponding to the uncompressed signal, wherein the encoded signal includes universally quantized dithered linear measurements of the signal, and wherein each universally quantized dithered linear measurement is a quantized dithered linear measurement of the signal missing one or more significant bits. Side information about the signal is obtained, and the side information is used to obtain a prediction of the signal. The missing one or more significant bits are determined from the encoded signal using the prediction of the signal and the missing one or more significant bits are combined with the encoded signal to produce quantized dithered linear measurements of the signal. Then, the signal can be reconstructed as a reconstructed signal using the quantized dithered linear measurements.

First claim

We claim:

1. A method for reconstructing a signal, wherein the signal is uncompressed, comprising steps of:
obtaining an encoded signal corresponding to the signal, wherein the encoded signal includes universally quantized dithered linear measurements of the signal, wherein each universally quantized dithered linear measurement is a quantized dithered linear measurement of the signal missing one or more significant bits;
obtaining side information about the signal;
using the side information to obtain a prediction of the signal;
determining the missing one or more significant bits from the encoded signal using the prediction of the signal;
combining the missing one or more significant bits with the encoded signal to produce quantized dithered linear measurements of the signal; and
reconstructing the signal as a reconstructed signal using the quantized dithered linear measurements, wherein the steps are performed in a decoder.

2. The method of claim 1 , wherein the side information is encoded as encoded side information, and further comprising:
decoding the encoded side information to obtain decoded side information.

3. The method of claim 1 , wherein the determining the one or more significant bits further comprises:
measuring the prediction of the signal using a linear operator to obtain linear measurements;
dithering the linear measurements to obtain dithered linear measurements;
quantizing the dithered linear measurements to obtain the quantized dithered linear measurements; and
extracting the missing one or more significant bits of each quantized linear measurement as the missing one or more significant bits.

4. The method of claim 1 , wherein the prediction of the signal is obtained from the side information using linear prediction.

5. The method of claim 1 further comprising:
correcting the missing one or more significant bits using the side information.

6. The method of claim 1 , further comprising:
updating the prediction of the signal and the side information using the reconstructed signal; and repeating the determining, the combining, and the reconstructing steps using the updated prediction and side information.

7. The method of claim 1 , wherein the side information includes a signal similar to the encoded signal.

8. The method of claim 1 , wherein the side information includes statistical similarity measures of the encoded signal with a signal similar to the encoded signal.

9. The method of claim 8 , wherein the statistical similarity measures includes a mean, a covariance, and a correlation of the signal.

10. The method of claim 1 , wherein the side information includes corrections to errors produced by determining the one or more missing significant bits.

11. The method of claim 10 , wherein the corrections are determined at an encoder.

12. The method of claim 1 , wherein the side information includes quantized measurements of the signal.

13. The method of claim 1 , wherein the signal is encoded in an encoder, and further comprising:
obtaining universal linear measurements of the signal;
adding dither to the universal linear measurement to produce dithered linear measurements; and
applying non-monotonic quantization to the universal linear measurements to produce the encoded signal.

14. The method of claim 13 , wherein the non-monotonic quantization further comprises:
quantizing the dither linear measurements using a monotonic quantizer; and
discarding the one or more significant bits of each measurement.

15. The method of claim 1 wherein the reconstructing solves an optimization problem.

16. The method of claim 15 , wherein the optimization enforces a sparsity model.

17. The method of claim 15 , wherein the optimization enforces low signal total variation.

18. The method of claim 15 , wherein the optimization uses weights derived from the side information.

19. The method of claim 15 wherein the optimization uses weights derived from the signal prediction.

20. The method of claim 15 , wherein the optimization is performed using a greedy algorithm.

21. The method of claim 15 , wherein the optimization is performed using a convex relaxation.

22. The method of claim 15 , wherein the optimization is using a dictionary.

23. The method of claim 22 , wherein the dictionary is learned from the prediction.

24. The method of claim 22 , wherein the dictionary is learned from the reconstructed signal.

25. The method of claim 1 , wherein the signal is a multispectral image.

26. The method of claim 1 , wherein the signal is acquired by a radar system.

27. The method of claim 26 , wherein the radar system is a synthetic aperture radar (SAR).

Assignees

• Mitsubishi Electric Res Laboratories Inc

Inventors

• Boufounos Petros T
• Valsesia Diego

Classifications

• H04N19/124 Primary

  Quantisation · CPC title

• G06T1/0007

  Image acquisition · CPC title

• G06V20/13

  Satellite images · CPC title

• H04N19/395

  involving distributed video coding [DVC], e.g. Wyner-Ziv video coding or Slepian-Wolf video coding · CPC title

• G06T2207/10036

  Multispectral image; Hyperspectral image · CPC title