Kernel Prediction based Blind Motion Deblurring
Latikesh Ahire, Aaryak Shah, Kumar Lakshya, Alok Kumar Kamal · 2024
This conference paper provides a dual-focus examination of image deblurring. The first part offers a comprehensive review, categorizing deblurring approaches into blind and non-blind methods for scenarios with limited or known blur information. The second part highlights recent deep learning advancements, emphasizing their transformative impact on deblurring tasks. Departing from conventional end-to-end approaches, it employs a non uniform motion blur prediction followed by non blind deconvolution block. Evaluation results demonstrate accurate motion blur estimates and competitive or superior restorations of real blurred images compared to existing methods. This paper contributes a concise yet comprehensive overview, bridging theoretical foundations with practical applications in the evolving landscape of image deblurring.