Efficient Video Deblurring Guided by Motion Magnitude and Convolutional Block Attention Module

Video deblurring is a pivotal task in the fields of low-level vision and graphics, aiming to restore clear videos from blurry sequences. The traditional approaches usually involve restoring the blurry middle frame by fusing multiple pixels. However, these methods overlook the varying degrees of blur present in each pixel of the video frame. This oversight results in unsatisfactory deblurring outcomes when incorporating blurry pixels into the video segment. To address this limitation, this paper proposes a framework that leverages the motion magnitude prior (MMP) as a guide for effective deep video deblurring. Moreover, we integrate a convolutional block attention module (CBAM) to tackle the issue of limited computing resources. By detecting the degree of pixel blur in each frame, the MMP is seamlessly incorporated into the recursive neural network for video deblurring. The CBAM introduces both channel and spatial attention, enabling a sequential attention structure from channel to space. By highlighting the critical information and neglecting unrelated extraneous details significantly enhances task processing efficiency and accuracy.