Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Weighting Map for Bit-Depth Expansion within a Rational Range

Yuqing Liu, Qi Jia, Jian Zhang, Xin Fan, Shanshe Wang, Siwei Ma, Wen Gao

TL;DR

Experimental results show the bit restoration network (BRNet) can restore colorful images with fewer artifacts and false contours, and outperforms state-of-the-art methods with higher PSNR/SSIM results and lower Wasserstein distance.

Abstract

Bit-depth expansion (BDE) is one of the emerging technologies to display high bit-depth (HBD) image from low bit-depth (LBD) source. Existing BDE methods have no unified solution for various BDE situations, and directly learn a mapping for each pixel from LBD image to the desired value in HBD image, which may change the given high-order bits and lead to a huge deviation from the ground truth. In this paper, we design a bit restoration network (BRNet) to learn a weight for each pixel, which indicates the ratio of the replenished value within a rational range, invoking an accurate solution without modifying the given high-order bit information. To make the network adaptive for any bit-depth degradation, we investigate the issue in an optimization perspective and train the network under progressive training strategy for better performance. Moreover, we employ Wasserstein distance as a visual quality indicator to evaluate the difference of color distribution between restored image and the ground truth. Experimental results show our method can restore colorful images with fewer artifacts and false contours, and outperforms state-of-the-art methods with higher PSNR/SSIM results and lower Wasserstein distance. The source code will be made available at https://github.com/yuqing-liu-dut/bit-depth-expansion

Learning Weighting Map for Bit-Depth Expansion within a Rational Range

TL;DR

Experimental results show the bit restoration network (BRNet) can restore colorful images with fewer artifacts and false contours, and outperforms state-of-the-art methods with higher PSNR/SSIM results and lower Wasserstein distance.

Abstract

Bit-depth expansion (BDE) is one of the emerging technologies to display high bit-depth (HBD) image from low bit-depth (LBD) source. Existing BDE methods have no unified solution for various BDE situations, and directly learn a mapping for each pixel from LBD image to the desired value in HBD image, which may change the given high-order bits and lead to a huge deviation from the ground truth. In this paper, we design a bit restoration network (BRNet) to learn a weight for each pixel, which indicates the ratio of the replenished value within a rational range, invoking an accurate solution without modifying the given high-order bit information. To make the network adaptive for any bit-depth degradation, we investigate the issue in an optimization perspective and train the network under progressive training strategy for better performance. Moreover, we employ Wasserstein distance as a visual quality indicator to evaluate the difference of color distribution between restored image and the ground truth. Experimental results show our method can restore colorful images with fewer artifacts and false contours, and outperforms state-of-the-art methods with higher PSNR/SSIM results and lower Wasserstein distance. The source code will be made available at https://github.com/yuqing-liu-dut/bit-depth-expansion
Paper Structure (13 sections, 10 equations, 11 figures, 6 tables, 1 algorithm)

This paper contains 13 sections, 10 equations, 11 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Relate Works
  3. Methodology
  4. Problem Formulation
  5. Network Design
  6. Progressive Training Strategy
  7. Quality Assessment
  8. Experiment
  9. Model Analysis
  10. Analysis on Progressive Training Strategy
  11. Analysis on Weighting Map Learning
  12. Comparison with State-of-the-Art Methods
  13. Conclusion

Figures (11)

  • Figure 1: Comparison between our method and a representative work RMFNet rmfnet_csvt2021 on restoring 2-bit image to 8-bit, showing our result on bottom part and the other on top part. RMFNet learns a direct mapping from low bit-depth to high bit-depth, that may change the given high-order bits, such as $01$ turns into $11$ in the instance. Our network learns a weight in a rational range and replenishes the missing bits without modifying the given information.
  • Figure 2: Network design of BRNet. An UNet-style architecture is designed to learn the weighting map for replenishing the LBD image to HBD.
  • Figure 3: Optimization-inspired block design. (a): Block for proximal operator (ProxBlock). (b): Block for RK-4 method (RK-4 Block). (c): OptBlock.
  • Figure 4: Loss comparison on progressive training (PT). The dash line denotes the converged loss value.
  • Figure 5: Two examples of PSNR/SSIM and W-Dis comparisons from restoring 4-bit to 8-bit.
  • ...and 6 more figures