LAPRAN: a scalable Laplacian pyramid reconstructive adversarial network for flexible compressive sensing reconstruction
View Patent ↗This disclosure addresses the single-image compressive sensing (CS) and reconstruction problem. A scalable Laplacian pyramid reconstructive adversarial network (LAPRAN) facilitates high-fidelity, flexible and fast CS image reconstruction. LAPRAN progressively reconstructs an image following the concept of the Laplacian pyramid through multiple stages of reconstructive adversarial networks (RANs). At each pyramid level, CS measurements are fused with a contextual latent vector to generate a high-frequency image residual. Consequently, LAPRAN can produce hierarchies of reconstructed images and each with an incremental resolution and improved quality. The scalable pyramid structure of LAPRAN enables high-fidelity CS reconstruction with a flexible resolution that is adaptive to a wide range of compression ratios (CRs), which is infeasible with existing methods.
1. A method for reconstructing a compressed image, comprising:
receiving first compressed image data based on a source image;
producing a first reconstruction of the source image at a first resolution from a first reconstructive adversarial network (RAN) using the first compressed image data as input to the first RAN;
receiving second compressed image data based on the source image; and
producing a second reconstruction of the source image at a second resolution from a second RAN using the first reconstruction and the second compressed image data as inputs to the second RAN.
2. The method of claim 1 , wherein producing the first reconstruction comprises:
extracting first features from the first compressed image data using a convolutional layer; and
using a fully connected layer to generate the first reconstruction from the first features.
3. The method of claim 2 , wherein producing the second reconstruction comprises:
generating an upscaled image from the first reconstruction;
extracting second features from the second compressed image data using a convolutional layer; and
generating an image residual to compensate for artifacts of the upscaled image using the second features.
4. The method of claim 3 , wherein producing the second reconstruction further comprises convolving the upscaled image and the image residual to produce the second reconstruction.
5. The method of claim 1 , further comprising:
training the first RAN; and
training the second RAN using an output of the trained first RAN.
6. The method of claim 5 , wherein training the first RAN uses a pixel-wise L 2 reconstruction loss and an adversarial loss.
7. The method of claim 1 , further comprising randomly encoding the source image to produce the first compressed image data at a first compression ratio and the second compressed image data at a second compression ratio.
8. The method of claim 7 , wherein the second compressed image data comprises the first compressed image data and additional information.