Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Edge-based Denoising Image Compression

Ryugo Morita, Hitoshi Nishimura, Ko Watanabe, Andreas Dengel, Jinjia Zhou

TL;DR

This work proposes a novel compression model that incorporates a denoising step with diffusion models, significantly enhancing image reconstruction fidelity by sub-information from leveraging latent space, offering a robust solution to the current limitations of image compression.

Abstract

In recent years, deep learning-based image compression, particularly through generative models, has emerged as a pivotal area of research. Despite significant advancements, challenges such as diminished sharpness and quality in reconstructed images, learning inefficiencies due to mode collapse, and data loss during transmission persist. To address these issues, we propose a novel compression model that incorporates a denoising step with diffusion models, significantly enhancing image reconstruction fidelity by sub-information(e.g., edge and depth) from leveraging latent space. Empirical experiments demonstrate that our model achieves superior or comparable results in terms of image quality and compression efficiency when measured against the existing models. Notably, our model excels in scenarios of partial image loss or excessive noise by introducing an edge estimation network to preserve the integrity of reconstructed images, offering a robust solution to the current limitations of image compression.

Edge-based Denoising Image Compression

TL;DR

This work proposes a novel compression model that incorporates a denoising step with diffusion models, significantly enhancing image reconstruction fidelity by sub-information from leveraging latent space, offering a robust solution to the current limitations of image compression.

Abstract

In recent years, deep learning-based image compression, particularly through generative models, has emerged as a pivotal area of research. Despite significant advancements, challenges such as diminished sharpness and quality in reconstructed images, learning inefficiencies due to mode collapse, and data loss during transmission persist. To address these issues, we propose a novel compression model that incorporates a denoising step with diffusion models, significantly enhancing image reconstruction fidelity by sub-information(e.g., edge and depth) from leveraging latent space. Empirical experiments demonstrate that our model achieves superior or comparable results in terms of image quality and compression efficiency when measured against the existing models. Notably, our model excels in scenarios of partial image loss or excessive noise by introducing an edge estimation network to preserve the integrity of reconstructed images, offering a robust solution to the current limitations of image compression.
Paper Structure (11 sections, 4 equations, 4 figures, 1 table)

This paper contains 11 sections, 4 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methods
  4. Edge Estimation Network(EEN)
  5. Canny-based Denosing Step(CDS)
  6. Temporary Data Complement Step
  7. Loss function and training
  8. Experiment
  9. Image compression Result
  10. Temporary Data Complement
  11. Conclusion

Figures (4)

  • Figure 1: The overview of our proposed model.
  • Figure 2: Qualitative comparison of the proposed method and the existing methods with smallest the bbp in each methods. Our model has a very high foreground image reconstruction system due to the acquisition of the edge information. It indicates the contours of the tree and castle are so sharp and refine compare with the existing model.
  • Figure 3: Quantitative result comparison of the proposed method and the existing methods on the DIV2L dataset. Our model outperforms the existing diffusion image compression model. F-PSNR and F-SSIM indicate the PSNR and SSIM for the foreground images. These images were masked by using a one-hot segmentation map detected by deeplabv3chen2018encoder.
  • Figure 4: Temporary Data Complement of the proposed method and other existing methods when the vector or latent information misses the region while transmission. “Inpainting" indicates the inpainting with diffusion model after receiving the reconstructed image. To compare with ours, the inpainting cannot consider the latent and edge information to complete the image, in contrast, ours can do it. The existing model doesn't have this solution for completing the image in missed the information.