Deep motion blur removal using noisy/blurry image pairs

Removing spatially variant motion blur from a blurry image is challenging problem as can be complicated and difficult to model accurately. Recent progress in deep neural networks suggests that kernel-free single deblurring achieved, but questions about performance persist. To improve performance, we proposed convolutional network restore sharp noisy/blurry pair captured quick succession. Two structures, Deblur Long Short-Term Memory (LSTM) DeblurMerger, are presented fuse the of images either sequential or parallel manner. boost training, gradient loss, adversarial spectral normalization leveraged. The training dataset consists pairs corresponding ground truth synthesized based on benchmark GOPRO. We evaluated trained variety synthetic datasets real pairs. results demonstrate approach outperforms state-of-the-art methods both qualitatively quantitatively. DeblurLSTM achieves best debluring while DeblurMerger nearly same result with significantly less computation time.