Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multiple Latent Space Mapping for Compressed Dark Image Enhancement

Yi Zeng, Zhengning Wang, Yuxuan Liu, Tianjiao Zeng, Xuhang Liu, Xinglong Luo, Shuaicheng Liu, Shuyuan Zhu, Bing Zeng

TL;DR

This work addresses compressed dark image enhancement by shifting the restoration problem from image space to latent space. It introduces two multi level VAEs to form parallel latent spaces for compressed and normal images, and a two branch latent space mapping network to translate compressed latent features into normal light latent representations. The approach yields state of the art results in both qualitative and quantitative evaluations across multiple datasets and quality factors, while maintaining texture details and suppressing blocking artifacts. The method also improves downstream tasks such as dark face detection, demonstrating meaningful practical impact for compressed image pipelines.

Abstract

Dark image enhancement aims at converting dark images to normal-light images. Existing dark image enhancement methods take uncompressed dark images as inputs and achieve great performance. However, in practice, dark images are often compressed before storage or transmission over the Internet. Current methods get poor performance when processing compressed dark images. Artifacts hidden in the dark regions are amplified by current methods, which results in uncomfortable visual effects for observers. Based on this observation, this study aims at enhancing compressed dark images while avoiding compression artifacts amplification. Since texture details intertwine with compression artifacts in compressed dark images, detail enhancement and blocking artifacts suppression contradict each other in image space. Therefore, we handle the task in latent space. To this end, we propose a novel latent mapping network based on variational auto-encoder (VAE). Firstly, different from previous VAE-based methods with single-resolution features only, we exploit multiple latent spaces with multi-resolution features, to reduce the detail blur and improve image fidelity. Specifically, we train two multi-level VAEs to project compressed dark images and normal-light images into their latent spaces respectively. Secondly, we leverage a latent mapping network to transform features from compressed dark space to normal-light space. Specifically, since the degradation models of darkness and compression are different from each other, the latent mapping process is divided mapping into enlightening branch and deblocking branch. Comprehensive experiments demonstrate that the proposed method achieves state-of-the-art performance in compressed dark image enhancement.

Multiple Latent Space Mapping for Compressed Dark Image Enhancement

TL;DR

This work addresses compressed dark image enhancement by shifting the restoration problem from image space to latent space. It introduces two multi level VAEs to form parallel latent spaces for compressed and normal images, and a two branch latent space mapping network to translate compressed latent features into normal light latent representations. The approach yields state of the art results in both qualitative and quantitative evaluations across multiple datasets and quality factors, while maintaining texture details and suppressing blocking artifacts. The method also improves downstream tasks such as dark face detection, demonstrating meaningful practical impact for compressed image pipelines.

Abstract

Dark image enhancement aims at converting dark images to normal-light images. Existing dark image enhancement methods take uncompressed dark images as inputs and achieve great performance. However, in practice, dark images are often compressed before storage or transmission over the Internet. Current methods get poor performance when processing compressed dark images. Artifacts hidden in the dark regions are amplified by current methods, which results in uncomfortable visual effects for observers. Based on this observation, this study aims at enhancing compressed dark images while avoiding compression artifacts amplification. Since texture details intertwine with compression artifacts in compressed dark images, detail enhancement and blocking artifacts suppression contradict each other in image space. Therefore, we handle the task in latent space. To this end, we propose a novel latent mapping network based on variational auto-encoder (VAE). Firstly, different from previous VAE-based methods with single-resolution features only, we exploit multiple latent spaces with multi-resolution features, to reduce the detail blur and improve image fidelity. Specifically, we train two multi-level VAEs to project compressed dark images and normal-light images into their latent spaces respectively. Secondly, we leverage a latent mapping network to transform features from compressed dark space to normal-light space. Specifically, since the degradation models of darkness and compression are different from each other, the latent mapping process is divided mapping into enlightening branch and deblocking branch. Comprehensive experiments demonstrate that the proposed method achieves state-of-the-art performance in compressed dark image enhancement.
Paper Structure (33 sections, 19 equations, 12 figures, 7 tables, 2 algorithms)

This paper contains 33 sections, 19 equations, 12 figures, 7 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Conventional Methods
  4. Deep-learning Methods
  5. Artifacts Reduction
  6. The Proposed Method
  7. Overview
  8. Expand to Multiple Latent Spaces
  9. Learning Multiple Latent Spaces
  10. Intuition of Multi-level VAE
  11. Multi-level VAE Training
  12. Learning Latent Space Mapping
  13. Latent Space Mapping Network
  14. Intuition of Latent Space Mapping
  15. Latent Space Mapping Training
  16. ...and 18 more sections

Figures (12)

  • Figure 1: Zoom-in details of the enhanced compressed dark image. (I-a, II-a, III-a) show the original images. (I-b) shows the comparative result with enhancement method. (II-b) shows the comparative result with the combination of the enhancement method and deblocking method (enhancement+deblocking). (III-b) shows the comparative result with the combination of deblocking method and enhancement method (deblocking+enhancement). ZeroDCEguo2020zero is a enhancement method and QGCNli-2020 is a deblocking method.
  • Figure 2: Conceptual comparison of two enhancement mechanisms. Image space enhancement enhances from image to image directly, and latent space enhancement work through mapping points on manifolds.
  • Figure 3: Mean Square Error (MSE) between compressed data and uncompressed data on dark face test dataset with 6000 dark images.
  • Figure 4: Intuition of multiple latent space mapping.
  • Figure 5: Framework of the proposed multi-level VAEs and multiple latent space mapping.
  • ...and 7 more figures