Improved Image Deblurring Algorithm to Generate Antagonistic Neural Network
Zou Qianying, Lan Luo, Wang Xiaofang, Chen Dongxiang, Liu Feng-yu · 2020
Aiming at the difficulty of the de-blurring kernel, poor de-blurring effect, and ringing caused by image noise in the traditional de-blurring method of moving image with the fuzzy kernel, a new algorithm based on edge discrimination and antagonistic neural network is proposed in this paper. The improved algorithm is based on the traditional condition to generate the antagonistic neural network by reducing the noise of the blurred image, extracting the edge information of the image and using it as the auxiliary information to guide the de-blurring effect in the generator. The discriminator network is improved, that is, two discriminators are used to distinguish the de-blurred image and the edge image extracted from the de-blurred image, and the two discriminators must be deceived simultaneously to achieve the de-blurring effect. Compared with the end-to-end deblurring algorithm in recent years, in the image dataset with 0.007 Gaussian noise, the PSNR of the improved algorithm is increased by 7.7%and the structural similarity is increased by 10% on average, but its running time is unchanged. Experimental results show that the improved algorithm can effectively remove the blurred image, and can be widely used in the field of natural motion-blurred image de-blurring.