用数据驱动方法解决声学相机图像去噪与拼接难题
A Data-Driven Paradigm-Based Image Denoising and Mosaicking Approach for High-Resolution Acoustic Camera
- 基于数据驱动范式设计去噪与拼接算法
- 在复杂噪声和窄视场下仍能有效恢复检测场景
- 适合水下成像、高分辨率声学图像处理领域
本文提出一种基于数据驱动范式的声学相机图像去噪与拼接方法。声学相机（即2D前视声呐）可在黑暗浑浊水域获取高分辨率声学图像，但因其独特的传感器成像机制，主流视觉处理方法如去噪与拼接仍处于初级阶段。由于声学图像中存在复杂噪声干扰且相机视场狭窄，即便收集大量图像也难以完整还原检测场景。现有研究多依赖先验知识与传感器模型设计手工算子，但此类方法易受噪声影响，特征细节不足。本研究提出一种数据驱动的声学图像去噪与拼接方法，并利用实测声学相机图像进行实验验证，结果表明该方法具有显著有效性。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。