Research Degraded Image Restoration Methods

Image is an important tool for human perceiving the existence of the world, not only can express the physical objects but also can make people understand the relationship between the recognizable objects much more conveniently, so that human could more profoundly study the physical world. The images can be acquired from the objective world by different observation systems and various means, however, there are all kinds of adverse factors during the image capture, such as various noises, motion blurring, atmospheric turbulence blurring, defocus blurring and so on. The images can be degraded, and then lose the worthy of application for real environment. Therefore, it is the very important value that the original scene can be restored and the quality the images can be improved.In this thesis, we have established several degraded image restoration models and designed the related algorithms to filter noise, motion blurring, atmospheric turbulence blurring and defocus blurring from the degraded images, and have studied the performance of the methods in terms of the image restoration quality and image detail preservation. The main content and innovation points are listed as follows:(1) To address the problem of different types of noise, by which the images are corrupted when they are captured, stored, converted and transmited, a new method based on Chebyshev theory and Radon transform is proposed for filtering the single noise or mixed noise from the corrupted images, PA for short. This method PA, where the parameters are identified adaptively in the related model, contains two stages:noise detection and noise reduction, which would remove impulse noise, Gaussian noise and other various noises mixed by impulse noise and Gaussian noise in different proportions. Extensive experimental results show that the proposed method PA outperforms the other classical methods in terms of subjective and objective image quality assessment, and demonstrate that the proposed model and algorithm PA is an effective image restoration method in the image denoising and image detail preservation.(2) To address the problem of motion blurring in the process of taking a image, which is mainly caused by the relative motion between the subject and the imaging equipment, a new method based on reaction-diffusion equation is proposed for restoring images that are degraded by motion blurring, RDER for short. This method RDER focuses on constructing the image restoration model and designing the related algorithm for filtering motion blurring, which takes advantage of motion blurring image quantization function and characteristics of reaction-diffusion equation about restoring image and preserve edge information. In particular, we introduce the image restoration capability of the RDER by model analysis and mathematical theory. To demonstrate the advantages of the RDER, which is compared with some common deblurring methods, extensive experimental results show that the proposed method RDER outperforms the other classical methods in terms of subjective and objective image quality assessment, and demonstrate that the proposed model and algorithm RDER is an effective image restoration method in the image motion deblurring and image detail preservation, meanwhile, has a certain robustness to noise interference.(3) To address the problem of image blurring caused by the atmospheric turbulence effect, a new method based on the image similarity and reaction-diffusion equation is proposed for restoring images that are degraded by the atmospheric turbulence blurring, ISRDE for short. The main task of this method ISRDE is to construct the image restoration model and design the related algorithm for filtering atmospheric turbulence blurring, which takes advantage of characteristics of optical similarity function and reaction-diffusion equation (RDE) about restoring image. In particular, several real remote sensing images are used in comparative experiment of algorithm to improve the application value of algorithm. Extensive experimental results show that the proposed method ISRDE outperforms the other classical methods in terms of subjective and objective image quality assessment, and demonstrate that the proposed model and algorithm ISRDE is an effective image restoration method in the image atmospheric turbulence deblurring and image detail preservation, meanwhile, has a certain robustness to noise interference.(4) To address the problem of image blurring caused by defocus, a new method based on the structure tensor and reaction-diffusion equation is proposed for restoring images that are degraded by the defocus blurring, STRDE for short. The main task of this method STRDE is to construct the image restoration model and design the related algorithm for removing defocus blurring, which takes full advantage of the geometric characteristics of the structure tensor, which can analys and process the structure information (for example, orientation field, edge, corner and so on), and reaction- diffusion equation (RDE) about restoring image. Extensive experimental results show that the proposed method STRDE outperforms the other classical methods in terms of subjective and objective image quality assessment, and demonstrate that the proposed model and algorithm STRDE is an effective image restoration method in the image defocus deblurring and image detail preservation, meanwhile, has a certain robustness to noise interference.(5) To address the problem of the degraded image restoration, combining the proposed methods of image denoising, motion deblurring, atmospheric turbulence deblurring and defocus deblurring, we have given the design of a degraded image restoration prototype system, which describes structure diagrams and general workflows of the prototype system based on the mathematical models, related algorithms and key technology. It is a simple and efficient solution, which is ease of implementation and has a wide range of application which is used in our daily life from military to civilian, from industry to commerce and so on.This thesis focuses on the research on degraded image restoration, where we have described the main factors causing image degradation, the categories of degraded factors and their mathematical models including various image noises, motion blurring, atmospheric turbulence blurring and defocus blurring. We have designed different degraded image restoration models and algorithms for removing different image degraded factors and improved the performance of these models in terms of the image restoration quality and image detail preservation. Finally, we have given an integrated solution for degraded image restoration, which is applied widely and translated into the product easily. In conclusion, the research works of this thesis give a new and useful attempt for degraded image restoration and will provide support and guarantee to the other researches in image processing.