Research of image deblurring based on the deep neural network

Image deblurring has a lot of applications which include but not limit recovery a high-quality Satellite imagery from a blurry one, extract more detail information from low resolution monitoring images, image compression and decompression. However, image deblurring has always been a difficult problem in the field of image processing. The traditional methods can effectively improve the image quality by image denoising technology. But this kind of methods has limit ability in recovery clear image from very vague images. Instead of previous methods, we focus on more challenger task that recovery high-resolution image from very vague images. We proposed to use the deep learning way to solve the problem and designed an end-to-end deep neural network for this task. To improve the quality of vague and ensure the accuracy of recovery images, we train our deep neural work by the adversarial training way and design our network based on encoder-decoder network using the convolution layer as network layers. The dataset we used to train our model is Celeba dataset. The results of our methods is promising compared with the traditional methods.