Restauração de imagens de sensoriamento remoto por métodos de deconvolução não linear restrita

Abstract

This work presents a comprehensive study on the deconvolution of orbital images, addressing its motivations, methodologies, and inherent challenges. Deconvolution is a fundamental technique for restoring image quality, particularly in satellite imagery of the Earth's surface, which is often degraded by atmospheric scattering, optical diffraction, and motion blur. Such distortions compromise the accuracy and utility of images in critical applications, including environmental monitoring, natural resource management, and disaster response. The methodology adopted in this research involves modeling the degradation mechanisms through the point spread function (PSF), implementing classical, iterative, and deep learning-based deconvolution algorithms, and conducting a comparative performance analysis. The evaluated methods include Inverse Filtering, Wiener filtering, Richardson-Lucy with and without "Total Variation" regularization, and "Deep Image Prior", under both known and blind PSF conditions. To support experimentation, the IDIA software tool was developed, integrating modules for image preprocessing and restoration, and metric-based image analysis. The algorithms were implemented and tested on synthetically degraded remote sensing images, with the objective of assessing each method’s effectiveness in enhancing spatial resolution, increasing contrast, and minimizing artifacts. The results demonstrate that iterative methods with regularization provide greater robustness against noise, while DIP offers stronger reconstruction capabilities without amplifying visual artifacts. Metric-based evaluation using PSNR, SSIM, NRMSE, and FWHM corroborates the visual assessments. The feasibility of this project was supported by the availability of advanced image processing tools and the growing demand for high-quality satellite imagery. Overall, this study contributes to the advancement of restoration techniques applied to Earth observation, offering a practical and extensible foundation for future developments.