Research On Adaptive Blur Image Restoration Algorithm

High quality images are eagerly demanded in deep space exploration, military reconnaissance, mapping, medical, civil and other fields which have been pursued by researchers all the time. When images are obtained from different observation system, camera image quality is degenerated from static design parameters because of many adverse factors such as atmospheric diffraction, defocus blur, motion blur and random noise. The application worth decreases at some extent. Therefore, research to improve image quality is particularly important. At present, image restoration has become an important method to obtain high-definition images in many fields.Point Spread Function(PSF) is a key parameter of image restoration. When PSF is unknown or evaluated inaccurately from degraded image, there is no doubt that image restoration becomes more difficult. The work on how to obtain exact PSF from single and mixed blur images is studied. The contributions of this paper are mainly as follows:(1) The mathematical model of optical imaging system is analyzed as well as common degradation model. Ill-posedness and how to improve the ill-posed problem in image restoration are simply discussed. This paper presents several image reconstruction algorithms and fast image deconvolution using Hyper-Laplacian priors which has the probability model more in line with the marginal distributions of real-world images. What's more, it has a better ringing inhibition effect.(2) To address the problem of Gaussian blur, an adaptive algorithm for PSF identification based on traditional knife-edge method is achieved. Image gradient criterion is designed based on Canny operator and Hough transform. Edge is extracted automatically on account of mean gray value difference between image blocks on both sides of the knife-edge. The restoration result of actual target image in terms of subjective and objective image quality assessment shows the effectiveness of the algorithm for Gaussian blur image.(3) For Gaussian blur image mixed with motion blur, it is necessary to pre-process the image to remove motion blur based on the frequency features. An improved method, which chooses the suitable sub-block in spectral image before edge detection based on Phase Congruency, is adopted to effectively avoid the problem of cross-shaped bright lines and errors introduced by smaller blur parameter. However, in consideration of serious noise in actual images, an interactive weak edge segmentation method called Grab Cut which could overcome brightness and contrast effect is adopted to restore remote sensing motion blur image of the moon after estimating blur parameter.(4) According to the study on Gaussian and motion blur image restoration in this paper, an adaptive algorithm for image restoration is designed. After evaluating the impact of motion blur parameter on knife-edge method, a motion parameter threshold TL is set accurately as a judgement of removing motion blur. In this algorithm, motion parameter L is firstly estimated from blur image. If L is greater than TL, the image is restored by knife-edge method after removing motion blur, otherwise it is feasible to restore blur image directly with the method of adaptive knife-edge.This paper focuses on the research of Gaussian blur image restoration by adaptive knife-edge method. Taking the universality and diversity of motion blur into account, it is necessary to set a motion parameter threshold TL which is used to design an adaptive restoration algorithm for the single and mixed blur image. This algorithm can not only obtain high-quality restored images, but also provides a convenient and valid solution for image restoration in various blur situations.