Research On Sharpening Methods For Low-quality Images

During the shooting process,the image is affected by weather(fog,rain,snow,cloudy),sports,insufficient light,etc.,combined with the limitations of the shooting hardware equipment,will cause serious degradation,resulting in loss of detail,reduced brightness,color degradation,blur,and recognition To reduce issues such as reduction,in order to improve the recognizability of the image and for subsequent detection,recognition,tracking,classification,etc.,this article focuses on the research on the clarity of low-quality images caused by noise and motion blur.A noise image clarification algorithm based on the combination of anisotropic diffusion and three-dimensional block matching algorithm and a motion blur image clarification algorithm based on the DeblurGAN model were developed.The main contents are as follows:(1)Research on the clarity of noisy images.This paper proposes a method for clearing noise images based on the combination of anisotropic diffusion and BM3 D.First,the improved anisotropic diffusion algorithm is used to preprocess the image.The edge enhancement operator with eight-direction diffusion and the diffusion coefficient function of the hyperbolic tangent function are used to improve the convergence speed,and the edge enhancement operator can enhance the image Edge area and detail information;then,the pre-processed image is processed by the BM3 D algorithm,and the smooth area and the edge area are searched for similar blocks in different ways.The smooth area is searched in the horizontal and vertical directions,and the edge area is in the vertical and edge directions.Search to better obtain the structural information of the image;finally,simulation and experiment verify the clearing effect of this method on noisy images.The simulation results show that the method in this paper can better retain the texture details of the original image and effectively avoid the edge ringing effect of the BM3 D algorithm,and improve the subjective visual experience.The experimental results show that the peak signal-to-noise ratio index after processing by the algorithm in this paper is 4.4%higher than that of the BM3 D algorithm,and the structural similarity is 3% higher.(2)Research on the clarification of motion blurred images.An improved DeblurGAN image motion blur clearing method is proposed to solve the problem of the DeblurGAN algorithm's insufficient image edge detail restoration and the grid effect of the restored image.First,in the generator network,a multi-scale convolution kernel neural network is used for feature extraction,a cascaded hole convolution is added and a self-adapting normalization method is used;the discriminator uses the Patch GAN network.Secondly,the gradient image L1 loss is introduced and combined with the confrontation loss and the perceptual loss,it is used as a regular constraint for image deblurring.Finally,we will train on the Go Pro dataset to get a deblurred model.The experimental results show that the method in this paper has a better subjective clarification effect and eliminates the grid effect;in addition,the peak signal-to-noise is 5.4% higher than the DeblurGAN algorithm,and the structural similarity index is 1% higher.