An Ensemble Approach to Motion Blur Reduction Using Weight Assignment

Abstract

Motion blur has been recognized as a significant factor contributing to performance degradation in the field of computer vision. In this paper, we introduce an innovative approach in deep learning to address the challenge of motion blur in autofocus, particularly in scenarios involving rapidly moving subjects. We employed an ensemble methodology that captured images at certain time intervals and assigned distinct weights based on image quality to effectively mitigate motion blur. This approach exhibited an average performance degradation of merely 1% when contrasted with exclusive training on high-quality images while yielding an average performance enhancement of over 3% when compared to the conventional method on both low-quality and high-quality images. © 2024 IEEE.