基于双向RNN和结构性剪枝的高效视频去模糊算法

urrent video deblurring algorithms often rely on explicit alignment of the input frames, but such alignment methods can yield inaccurate or even erroneous inter-frame information when dealing with blurry videos. This misalignment leads to suboptimal deblurring results. Therefore, research on video deblurring algorithms that do not require explicit frame alignment holds significant academic value. In this paper, we address the aforementioned issue by proposing an efficient video deblurring algorithm based on Bidirectional Recurrent Neural Networks (RNNs) and structured pruning techniques. Our method leverages the hidden states of RNNs to capture and integrate inter-frame information without explicit alignment, thereby accomplishing motion compensation and fusion. A reconstruction module is further employed to generate sharp frames. Moreover, we introduce an improved structural pruning approach that effectively reduces the model\'s parameter count, thus enabling efficient generation of high-quality clear video frames. Extensive comparative experiments demonstrate that our method achieves state-of-the-art deblurring performance, while ablation studies validate the rationality and effectiveness of the proposed scheme.