Generic compression ratio adapter for end-to-end data-driven compressive sensing reconstruction frameworks
View Patent ↗A compression ratio (CR) adapter (CRA) for end-to-end data-driven compressive sensing (CS) reconstruction (EDCSR) frameworks is provided. EDCSR frameworks achieve state-of-the-art reconstruction performance in terms of reconstruction speed and accuracy for images and other signals. However, existing EDCSR frameworks cannot adapt to a variable CR. For applications that desire a variable CR, existing EDCSR frameworks must be trained from scratch at each CR, which is computationally costly and time-consuming. Embodiments described herein present a CRA framework that addresses the variable CR problem generally for existing and future EDCSR frameworks with no modification to given reconstruction models nor enormous additional rounds of training needed. The CRA exploits an initial reconstruction network to generate an initial estimate of reconstruction results based on a small portion of acquired image measurements. Subsequently, the CRA approximates full measurements for the main reconstruction network by complementing the sensed measurements with a re-sensed initial estimate.
1. A method for adapting a compressive sensing framework for a variable compression ratio, the method comprising:
generating a first reconstruction of an original signal based on a portion of first compressive measurements of the original signal; and
generating a second reconstruction of the original signal based on the first compressive measurements and second compressive measurements of the first reconstruction.
2. The method of claim 1 , wherein generating the second reconstruction of the original signal comprises generating the second reconstruction at a fixed resolution from the first compressive measurements having a variable compression ratio.
3. The method of claim 1 , further comprising generating the first compressive measurements by compressively sensing the original signal.
4. The method of claim 1 , further comprising generating the second compressive measurements by compressively sensing the first reconstruction.
5. The method of claim 4 , further comprising concatenating the first compressive measurements with the second compressive measurements;
wherein generating the second reconstruction comprises inputting the concatenated first compressive measurements and second compressive measurements into a machine learning algorithm.
6. The method of claim 1 , wherein the first reconstruction is generated using machine learning.
7. The method of claim 6 , wherein:
the first reconstruction of the original signal is generated with a first neural network; and
the second reconstruction of the original signal is generated with a second neural network.
8. The method of claim 1 , wherein the second reconstruction comprises a final reconstruction of the original signal.
9. The method of claim 1 , further comprising generating a third reconstruction of the original signal based on the first compressive measurements and third compressive measurements of the second reconstruction.
10. The method of claim 9 , wherein the third reconstruction comprises a final reconstruction of the original signal.
11. A compression ratio adapter, comprising:
a first reconstruction network configured to generate a first reconstruction of an original signal based on a portion of first compressive measurements of the original signal; and
a second reconstruction network configured to generate a second reconstruction of the original signal based on the first compressive measurements and second compressive measurements of the first reconstruction.
12. The compression ratio adapter of claim 11 , wherein the compression ratio adapter is configured to receive the first compressive measurements having a variable compression ratio.
13. The compression ratio adapter of claim 11 , further comprising a compressive sensor configured to produce the first compressive measurements of the original signal.
14. The compression ratio adapter of claim 11 , further comprising a compressive resensor configured to produce the second compressive measurements of the first reconstruction in accordance with an input size of the second reconstruction network.
15. The compression ratio adapter of claim 14 , further comprising a concatenator configured to concatenate the first compressive measurements with the second compressive measurements.
16. The compression ratio adapter of claim 11 , further comprising a third reconstruction network configured to generate a third reconstruction of the original signal based on the first compressive measurements and third compressive measurements of the second reconstruction.
17. The compression ratio adapter of claim 16 , further comprising a first compressive resensor configured to produce the second compressive measurements of the first reconstruction in accordance with an input size of the second reconstruction network.
18. The compression ratio adapter of claim 17 , further comprising a second compressive resensor configured to produce the third compressive measurements of the second reconstruction.
19. The compression ratio adapter of claim 11 , further comprising a third reconstruction network in parallel with the first reconstruction network and configured to generate a third reconstruction of the original signal based on the portion or another portion of the first compressive measurements of the original signal.
20. The compression ratio adapter of claim 11 , further comprising a third reconstruction network in parallel with the second reconstruction network and configured to generate a third reconstruction of the original signal based on the first compressive measurements and the second compressive measurements of the first reconstruction.