Image defocusing blurring method based on deep learning

Abstract

The invention belongs to the technical field of digital image intelligent processing, and particularly relates to an image defocusing blurring method based on deep learning. The method comprises the following steps: constructing a defocusing blurred data set by shooting or adding random blurring and the like, so that each group of data comprises a clear image as an original image and a plurality of blurred images as blurred images corresponding to the clear image; training a defocusing fuzzy deep neural network; recovering a blurred object out of a focal plane from the image through a deep neural network by using a non-alignment loss function; a non-pixel-level aligned deblurred data set is shot in a real scene, and a deep neural network is trained through a non-alignment loss function. Experimental results show that the out-of-focus blurred image shot in a real scene can be effectively recovered, and the proposed data set can effectively train the out-of-focus blurred network througha non-alignment loss function. The method can be used for camera zooming, robot vision systems and the like.

Description

technical field [0001] The invention belongs to the technical field of digital image intelligent processing, and in particular relates to an image deblurring method, and more specifically, relates to an image defocus blurring method based on deep learning. Background technique [0002] In recent years, with the rapid development of deep learning, blind image deblurring (Blind image deblurring) has made some progress: Orest et al. proposed an end-to-end motion blurring algorithm [1] , achieved the highest objective quality evaluation (PSNR and SSIM) scores at that time, and good visual effects. This article is the first time to use the addition of GAN loss and content loss to design a loss function for motion blur, and conducted a controlled experiment to prove the effectiveness of the loss function; then Orest et al. improved the method [3] , introducing the Feature Pyramid Network as a generator. The speed and effect of the original method have been further improved; Tao ...

AI Technical Summary

Problems solved by technology

Since the camera focusing will change the image distance or focal length of the camera, there is no pixel-level alignment relationship between the clear image and the out-of-focus blurred image. If you use a pixel-level aligned loss function, such as L1 and MSE loss function, it will lead to defocusing and blurring. The border and outline of the picture after

Images