A Rule-based Optimized Mathematical Framework for High-fidelity Aerial Image Restoration

Abstract

Aerial image deblurring is a vital challenge in drone vision systems, where Gaussian, motion, and out-of-focus blurs significantly lowering the image quality and the quality of subsequent analysis. Standard deblurring algorithms usually implement one method uniformly in all situations without recognizing the type of blur. To address this issue, we introduce ROBIN (Rule, Based Optimization for Blur Identification and Neutralization), a hybrid deblurring system that identifies blur types and uses appropriate restoration methods. Our model uses a deep learning, based classifier to decide which of the three blur categories the input image belongs to. After the classification, a rule-based controller assigns each image to the respective restoration method: Total Variation, regularized Wiener deconvolution for Gaussian blur, Particle Swarm Optimization, augmented Richardson, Lucy deconvolution for motion blur, and Genetic Algorithm, tuned blind deconvolution for out of focus blur. Such a modular and adaptive system not only makes the process computationally efficient but also allows a high degree of restoration fidelity. A wide range of experiments under PSNR, SSIM, MSE, RMSE, and entropy metrics confirm that the proposed method is better than the traditional and state of the art methods in all blur categories in terms of clarity, detail retention, and robustness. ROBIN is an efficient solution for real-time aerial imaging use cases in environmental monitoring, surveillance, and disaster response missions. Experimental results show that the proposed method achieves a PSNR of 31.26 dB, SSIM of 0.94, and Entropy of 8.98, which is far superior to all the state-of-the-art approaches.