| Images taken outdoors are severely degraded by adverse factors,such as weather(haze,rain,snow,and cloud),motion,and insufficient light,etc..This not only affects the images visual effects,but also causes important details of the images to be lost,making their useful values to be greatly affected.Therefore,it is of great significance to improve the sharpness of these low-quality images affected by adverse factors.In this thesis,the highlights are placed on the clearing methods for low-quality images caused by haze weather and motion blur.The main work is as follows:(1)For image dehazing,the defogging method is applied to images taken in haze weather condition.Analyses and comparisons on several typical image dehazing algorithms are carried out,and then an image defogging algorithm for conditional generative adversarial networks is proposed.The new algorithm is tried and improved on the network structure and loss function.First,the new algorithm utilizes the Dense Net instead of the traditional U-net as the network structure of the generator,and uses Patch-GAN as the network structure of the discriminator.Secondly,its uses the pre-trained visual geometry group(VGG)model and the total variation regularization gradient to modify the loss function.The new algorithm is an end-to-end defogging algorithm.It does not need to estimate the projection map and related haze features to obtain a defogged image,which can be applied to a variety of scenarios.The algorithm effectively reduces the halo phenomenon and haze residue problem caused by the traditional method after dehazing,and can better preserve the details of the original image.The test results show that the improvement improves its structural similarity index from 76.9% to 93.4%.(2)For image deblurring,the clearing methods are applied to blurred images caused by the relative displacement between the shooting equipment and the target object.Based on the researched on several classical deblurring algorithms,an image deblurring mathematical model combining overlapping group sparse total variation and non-local total variation model is proposed.Firstly,the fuzzy image is preprocessed,that is,the image is denoised.Then the radon transform is used to calculate the motion kernel size of the blurred image.Finally,the mathematical model is used to reconstruct the image.The experimental results show that the proposed algorithm can clearly recover the blurred image and better retain the texture details of the image. |
