基于非局部自回归模型的遥感图像高保真抑噪方法研究

High-quality, high-resolution remote sensing imaging is in high demand by national economy and security. The introduction of mixed noise in the imaging process constitutes the major reason for resolution and quality reduction, which severely constricts the development of remote sensing imaging. Therefore, it is essential to effectively remove the noise for quality enhancement. In image denoising, the key issue is to remove noise effectively while retaining image detail information as much as possible. However, high-fidelity denoising is unattainable because existing method cannot model and represent remote sensing image structure feature precisely, which will induce reconstruction error during the restoration. To address this problem, this project will focus on effective image structure modeling in mixed noise circumstance and high-fidelity remote sensing image denoising. By exploiting image structure feature, robust remote sensing image model training methods will be studied. The major purpose of this project is to establish a high precise image structure model, which can be used in high-fidelity image noise removal. Through optimal image representation, the original image can be accurately rebuilt during denoising procedure and the image signal will be processed with high-fidelity. It is expected that this project will solve the problem of losing image detail information during the denoising procedure, which can provide theoretical foundation and technical support for the quality improvement of the high-resolution earth observation images.

高质量、高清晰度遥感成像在国家经济和安全等领域有重大需求。成像过程中引入的复杂噪声是导致图像质量下降的主要因素。如何在降噪过程中保持图像信息，实现信号高保真是遥感图像处理的难点。现有方法对遥感图像复杂结构特征的认知、建模和表达不精确，会在重建清晰信号时引起重构误差，造成图像信息的损失。为此，本课题拟围绕复杂噪声背景下的遥感图像结构自适应建模展开研究：(1)探索高分辨率对地观测遥感图像信号的结构特性；(2)构建具有较强鲁棒性的遥感图像结构自适应模型；(3)研究基于非局部自回归模型的遥感图像“高斯与脉冲”混合噪声抑制方法，进而在降噪过程中实现对原始图像的准确重建和信号高保真。预期通过本项目研究，为我国高分辨率对地观测图像质量的提升奠定理论基础，提供技术保障。

卫星遥感成像是对地观测的重要手段之一，在国家安全、国民经济、科学研究和人民生活等方面有重要作用。遥感卫星图像需要真实、客观的反映地物特征。然而，由于星载成像系统处于复杂的电磁环境中，卫星内部电路系统众多，CCD 器件的工艺缺陷等，在获取和传输图像的过程中会引入大量噪声，严重降低对地观测获取的图像质量，给后期的图像处理和人眼视觉判读带来障碍。因此，对遥感图像的噪声分析和处理是提高遥感图像质量和准确度的必经之路。本课题面向实际遥感成像系统中存在的“高斯与脉冲”混合噪声，研究高效、高保真的遥感图像降噪方法。通过探索高分辨率对地观测遥感图像信号的特性，研究基于结构自适应的遥感图像模型构建方法，实现对图像的准确刻画与优化表达，进而在降噪过程中降低图像信息的损失。