A Data-Driven Paradigm-Based Image Denoising and Mosaicking Approach for High-Resolution Acoustic Camera

TL;DR

This paper introduces a data-driven approach for denoising and mosaicking high-resolution acoustic camera images, addressing challenges posed by complex noise and limited field of view in underwater environments.

Contribution

It presents a novel data-driven method for acoustic image denoising and mosaicking, improving robustness over traditional handcrafted operator techniques.

Findings

Effective noise reduction demonstrated in experiments

Improved scene restoration from multiple acoustic images

Robustness against noise interference confirmed

Abstract

In this work, an approach based on a data-driven paradigm to denoise and mosaic acoustic camera images is proposed. Acoustic cameras, also known as 2D forward-looking sonar, could collect high-resolution acoustic images in dark and turbid water. However, due to the unique sensor imaging mechanism, main vision-based processing methods, like image denoising and mosaicking are still in the early stages. Due to the complex noise interference in acoustic images and the narrow field of view of acoustic cameras, it is difficult to restore the entire detection scene even if enough acoustic images are collected. Relevant research work addressing these issues focuses on the design of handcrafted operators for acoustic image processing based on prior knowledge and sensor models. However, such methods lack robustness due to noise interference and insufficient feature details on acoustic images. This study proposes an acoustic image denoising and mosaicking method based on a data-driven paradigm and conducts experimental testing using collected acoustic camera images. The results demonstrate the effectiveness of the proposal.