FrameDeblur: NAFNet-Based Selective Video Frame Deblurring and Restoration Framework

Abstract: This paper presents FrameDeblur, a video-based image restoration framework designed to identify and deblur individual blurred frames from a video sequence. Unlike conventional image deblurring approaches that directly operate on a single image or process an entire video sequence, the proposed system accepts a video as input and provides an interactive mechanism for manually selecting the frame requiring restoration. The selected frame is extracted and subjected to preprocessing before being passed through a pretrained NAFNet (Nonlinear Activation Free Network) for deep image restoration. The restored output is subsequently refined using classical image restoration and enhancement operations to improve sharpness, structural details, and overall visual quality. The system integrates video frame extraction, manual frame navigation, deep learning-based restoration, and final image generation within a unified interactive workflow. By concentrating computational processing on the user-selected frame, the framework avoids unnecessary restoration of the complete video sequence and provides a practical approach for selective frame-level deblurring. The system is implemented with an interactive Gradio-based interface, enabling users to upload a video, navigate through its frames, select a blurred frame, and obtain the corresponding deblurred image as output. The proposed framework provides a flexible foundation for future extensions involving automatic blur detection, sequential video restoration, temporal consistency, and real-time deblurring.

Keywords: Video Deblurring, Frame Selection, Image Restoration, NAFNet, Deep Learning, Motion Blur, Video Processing, Frame Enhancement, Selective Restoration, Gradio.