High Quality Underwater Image Compression with Adaptive Correction and Codebook-based Augmentation

TL;DR

Underwater image compression is challenged by light attenuation, color casts, and limited bandwidth. The paper presents HQUIC, a framework that combines adaptive lighting/tone correction, a codebook-based auxiliary information path, and a frequency-aware transformer to exploit underwater-specific features for improved rate-distortion. Through extensive evaluations on diverse underwater datasets, HQUIC achieves state-of-the-art performance with substantial BD-rate gains and robust generalization across high- and low-quality imagery. This approach offers a practical solution for efficient underwater data transmission, with potential impact on autonomous underwater missions and marine monitoring.

Abstract

With the increasing exploration and exploitation of the underwater world, underwater images have become a critical medium for human interaction with marine environments, driving extensive research into their efficient transmission and storage. However, contemporary underwater image compression algorithms fail to fully leverage the unique characteristics distinguishing underwater scenes from terrestrial images, resulting in suboptimal performance. To address this limitation, we introduce HQUIC, designed to exploit underwater-image-specific features for enhanced compression efficiency. HQUIC employs an ALTC module to adaptively predict the attenuation coefficients and global light information of the images, which effectively mitigates the issues caused by the differences in lighting and tone existing in underwater images. Subsequently, HQUIC employs a codebook as an auxiliary branch to extract the common objects within underwater images and enhances the performance of the main branch. Furthermore, HQUIC dynamically weights multi-scale frequency components, prioritizing information critical for distortion quality while discarding redundant details. Extensive evaluations on diverse underwater datasets demonstrate that HQUIC outperforms state-of-the-art compression methods.

Figures (8)

• Figure 1: Left: Schematic Diagram of the Jaffe-McGlamery model. It explains the influence of illumination on underwater imaging, which is different from that of terrestrial images. Right: Under the special lighting conditions underwater, compression models designed for terrestrial images struggle with detail reconstruction, while the method we propose can effectively handle the impacts brought about by the underwater lighting conditions.
• Figure 2: The three channel intensity distribution across various datasets, including terrestrial image datasets (CLIC2020 and DIV2k) represented in the left, and underwater image datasets (USOD10k and SUIM) depicted in the right, demonstrates that underwater images exhibit lower intensity values in the red channel and higher intensity values in blue and green channels compared to terrestrial images.
• Figure 3: Overview of the proposed methods. (a) presents the overall architecture of the pipeline, and (b)-(d) present the detailed modules. The original image undergoes modulation (\ref{['sec:ALTC']}) and encoding using two distinct branches, $E$ and $E_{vq}$, which consist of Weighted Frequency Block (WFB)(\ref{['sec:fwt']}) and resnet block respectively. The resulting compressed features, along with modulation coefficients, are then transmitted. To facilitate information exchange between $D$ and $D_{vq}$, Adaptive Information Enhancement (AIE) modules are employed(\ref{['sec:cb']}), enabling the auxiliary branch to augment the main branch's information. Finally, the decompressed feature is demodulated, producing the reconstructed image.
• Figure 4: Three channel feature distributions after adaptive light and tone correction (ALTC) on the SUIM dataset. As seen in the plot, the mean of red channel is nearing 0, which is helpful for the subsequent models. Compared to the distribution in \ref{['pic:violin']}, ALTC can make it easy for the model to train.
• Figure 5: The detailed architecture of Weighted Frequency Block (WFB) and Frequency-based Weighted Transformer (FBWT) Module. Standard Transformer represents the widely used transfromer architecture in vision task.