Image motion blur removing method based on improved cyclic generative adversarial network

Abstract

The invention discloses an image motion blur removing method based on an improved cyclic generative adversarial network. The method comprises the following steps: 1, constructing a non-paired blur clear data set; 2, constructing a generator network composed of an encoder, a feature converter and a decoder; 3, constructing a discriminator network for dividing the image by a receptive field; 4, constructing a joint loss function; 5, constructing two mirrored annular GAN networks to obtain a cyclic generative adversarial network model; 6, inputting a motion blurred image to be processed into the trained model in the step 5 to obtain a deblurred image; 7, carrying out two-dimensional Fourier transform on the preliminary deblurred image obtained in the step 6, and filtering out high-frequency bright spot spectrum information to obtain an accurate clear image. According to the method, a fuzzy kernel does not need to be estimated, calculation parameters are few, the deblurring speed is high, the problems of mode collapse and gradient disappearance are avoided, and the false recognition problem of frequency domain pseudo-clearness is solved.

Description

technical field [0001] The invention belongs to the technical field of computer image processing, and relates to an image motion blur removal method based on an improved cycle generation confrontation network. Background technique [0002] Image deblurring has always been one of the important research directions in the field of computer vision and image processing. When shooting by hand, the motion blur caused by the relative movement between the camera lens and the scene during the exposure process becomes one of the important reasons affecting the recognition results. [0003] Motion blur is also called dynamic blur. According to whether the blur kernel is known, the image deblurring problem is mainly divided into blind deblurring algorithm and non-blind deblurring algorithm. The principle of non-blind image deblurring is to use known blur kernels to perform deconvolution operations to complete image restoration. In 2011, Zhao et al. [1] proposed a new image deblurring a...

AI Technical Summary

Problems solved by technology

The above methods still have problems such as inaccurate blur kernel estimation and lack of feature consistency loss constraints. They did not pay attention to the fact that the deblurred image will have redundant high-frequency spectral components in the frequency domain, resulting in the deblurred image. The original image produces new errors, which reduces the recognition accuracy of the image

Images