Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bit-depth color recovery via off-the-shelf super-resolution models

Xuanshuo Fu, Danna Xue, Javier Vazquez-Corral

TL;DR

The paper addresses the challenge of recovering high-bit-depth color representations from lower-bit-depth images, a problem prone to banding and texture loss. It introduces a framework that recovers color depth bit-by-bit by leveraging off-the-shelf super-resolution priors, using a multi-scale feature extractor to provide rich spatial context and an attention-augmented bit-plane predictor to refine each added bit-plane. Key contributions include (i) using frozen pre-trained SR encoders at multiple scales as priors, (ii) a multi-path feature extractor with CBAM fusion, and (iii) a lightweight, attention-enabled bit-plane prediction network that progressively reconstructs higher bit-depth images. Experiments on Sintel, TESTIMAGES, Kodak, and ESPL v2 demonstrate consistent improvements in PSNR and SSIM over state-of-the-art methods, underscoring the practical impact of SR-informed priors for high-fidelity color restoration.

Abstract

Advancements in imaging technology have enabled hardware to support 10 to 16 bits per channel, facilitating precise manipulation in applications like image editing and video processing. While deep neural networks promise to recover high bit-depth representations, existing methods often rely on scale-invariant image information, limiting performance in certain scenarios. In this paper, we introduce a novel approach that integrates a super-resolution architecture to extract detailed a priori information from images. By leveraging interpolated data generated during the super-resolution process, our method achieves pixel-level recovery of fine-grained color details. Additionally, we demonstrate that spatial features learned through the super-resolution process significantly contribute to the recovery of detailed color depth information. Experiments on benchmark datasets demonstrate that our approach outperforms state-of-the-art methods, highlighting the potential of super-resolution for high-fidelity color restoration.

Bit-depth color recovery via off-the-shelf super-resolution models

TL;DR

The paper addresses the challenge of recovering high-bit-depth color representations from lower-bit-depth images, a problem prone to banding and texture loss. It introduces a framework that recovers color depth bit-by-bit by leveraging off-the-shelf super-resolution priors, using a multi-scale feature extractor to provide rich spatial context and an attention-augmented bit-plane predictor to refine each added bit-plane. Key contributions include (i) using frozen pre-trained SR encoders at multiple scales as priors, (ii) a multi-path feature extractor with CBAM fusion, and (iii) a lightweight, attention-enabled bit-plane prediction network that progressively reconstructs higher bit-depth images. Experiments on Sintel, TESTIMAGES, Kodak, and ESPL v2 demonstrate consistent improvements in PSNR and SSIM over state-of-the-art methods, underscoring the practical impact of SR-informed priors for high-fidelity color restoration.

Abstract

Advancements in imaging technology have enabled hardware to support 10 to 16 bits per channel, facilitating precise manipulation in applications like image editing and video processing. While deep neural networks promise to recover high bit-depth representations, existing methods often rely on scale-invariant image information, limiting performance in certain scenarios. In this paper, we introduce a novel approach that integrates a super-resolution architecture to extract detailed a priori information from images. By leveraging interpolated data generated during the super-resolution process, our method achieves pixel-level recovery of fine-grained color details. Additionally, we demonstrate that spatial features learned through the super-resolution process significantly contribute to the recovery of detailed color depth information. Experiments on benchmark datasets demonstrate that our approach outperforms state-of-the-art methods, highlighting the potential of super-resolution for high-fidelity color restoration.
Paper Structure (12 sections, 2 equations, 2 figures, 5 tables)

This paper contains 12 sections, 2 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overview
  4. SR-based Multi-scale Feature Extractor
  5. Attention-augmented Bit-plane Prediction Network
  6. Experiments
  7. Datasets
  8. Implementation Details
  9. Quantitative Results
  10. Qualitative Results
  11. Ablation Study
  12. Conclusion

Figures (2)

  • Figure 1: The framework of our approach. The bit-depth recovery model comprises several submodels with the same architecture but different weights, and recovers the color depth bit-by-bit with each submodel. Given a b-bit image, the submodel predicts three binary bit-planes for R, G, and B channels, respectively. The predicted bit-plane is concatenated with the binary bit-planes of the input image and then mapped back to the b+1-bit values. By processing the image step by step, we can obtain the final target image. Each submodel includes two parts: (a) the multi-scale feature encoder and (b) the bit-plane prediction network. The multi-scale feature encoder consists of several super-resolution encoders pretrained for different scale SR tasks, an inception module, and a feature aggregation module that fuses the features. The multi-scale features are then processed by the bit-plane prediction network including different blocks to predict the output binary bit-planes.
  • Figure 2: Visual comparison of our method versus BitMore. From left to right: Ground truth, Bitmore and Ours. Rows 2 and 4 present the full images, while rows 1 and 3 are close-ups.