PAMSGAN: Pyramid Attention Mechanism-Oriented Symmetry Generative Adversarial Network for Motion Image Deblurring

Motion blur is a common problem in optical imaging, which caused by the relative displacement between subject and camera exposure process of camera. This can result motion acquired image, reduce image resolution affect imaging quality. restoration technology uses existing to restore clear through modeling physical mathematical solution without re-photographing target scene. It has an important application value civil military fields. Solving jitter object during very challenging problem. When popular generative adversarial network model directly applied blind removal task, serious pattern collapse phenomenon will occur. In this paper, we propose novel deblurring based on pyramid attention mechanism-oriented symmetry network. new method does not need predict fuzzy kernel blurred images, realize blur. Based original CycleGan, structure loss function are improved. The accuracy images improved, stability greatly enhanced case limited samples. adopts encoding decoding structure, introduces feature mechanism. combination multi-scale features mechanism capture more rich advanced improve performance. experiment, RMSProp algorithm used optimize training. Finally, obtained training discriminant Experimental results related benchmark datasets show that quality proposed higher terms subjective objective evaluation. Meanwhile, achieve better subsequent detection tasks.