Unified Residual Translation for Paired Image Restoration
Organizations: University of Delaware · BRAC University · Indiana University Indianapolis · University of Alabama at Birmingham
Abstract
Paired image restoration spans changes in imaging contrast, sensor modality, and spatial resolution. These settings differ in their input channels and target grids, raising the question of how much of the prediction pipeline must be specialized to each task. We present the Unified Residual Translator (URT), which separates fixed channel and scale mapping from learned residual prediction. A deterministic reference uses channel averaging, replication, or bicubic interpolation to match the target dimensions, while a NAFNet encoder-decoder predicts an additive correction from the complete input. Separately trained instances share the prediction rule, objective, and optimizer across MRI contrast synthesis, visible-to-infrared translation, SAR-to-optical translation, and real-camera super-resolution. Across six task and scale settings, URT achieves the highest PSNR and SSIM among the evaluated baselines on IXI, SEN1-2, and RealSR and . It reaches 30.774dB on IXI and 13.478dB on SEN1-2, with luminance PSNR of 33.596, 28.742, and 27.342dB at the three RealSR scales. On LLVIP, it ranks second in PSNR and SSIM. A study of four spectral alternatives identifies complementary benefits of fitted references and bandwise residual gains. Together, these results support explicit dimensional alignment and learned residual correction as a common formulation for heterogeneous paired imaging tasks.
Figures & tables
| Task | Reference | ||
|---|---|---|---|
| IXI | 1 | Five-slice mean | |
| LLVIP | 1 | Visible-channel mean | |
| SEN1-2 | 1 | SAR replicated to RGB | |
| RealSR | Bicubic RGB upsampling |
| Model | PSNR | SSIM | LPIPS | Params. |
|---|---|---|---|---|
| IXI | ||||
| URT (ours) | 30.774 | 0.957 | 0.032 | 38.292 |
| pix2pix [ 13 ] | 28.747 | 0.934 | 0.030 | 54.412 |
| SynDiff [ 25 ] | 22.154 | 0.700 | 0.117 | 39.725 |
| LLVIP | ||||
| URT (ours) | 13.126 | 0.359 | 0.569 | 17.111 |
| IXI | LLVIP | SEN1-2 | RealSR | |||||
|---|---|---|---|---|---|---|---|---|
| Model | PSNR | SSIM | PSNR | SSIM | PSNR | SSIM | PSNR-Y | SSIM-Y |
| URT (B0) | 30.774 | 0.957 | 13.126 | 0.359 | 13.478 | 0.229 | 33.596 | 0.920 |
| B1 | 30.677 | 0.940 | 12.958 | 0.353 | 13.667 | 0.228 | 33.467 | 0.914 |
| B2 | 30.663 | 0.939 | 13.355 | 0.366 | 13.492 | 0.236 | 33.489 | 0.915 |
| B2r | 30.629 | 0.938 | 13.332 | 0.377 | 13.407 | 0.235 | 29.451 | 0.758 |
| B3 | 30.389 | 0.937 | 13.317 | 0.382 | 13.520 | 0.236 | 33.260 | 0.909 |
Image Restoration
Image restoration uses computational methods to recover useful visual content from noisy, blurred, low-resolution, or otherwise degraded images.
See all 399 papers