A Common Fuzzy Image Restoration Research

Image restoration is a very important research field in the area of digital image processing. Due to the relative movement, defocus and noise, the grabbed image is not as good as we expected. This thesis focuses on the whole process of image restoration. Firstly, identify the motion blurred and defocus blurred images. Secondly, estimate the point spread function PSF (point spread function) of system. Finally, restore blurred images with improving winner filter and regularization method.Identify the two common blurred types:motion blurred and defocus blurred. After flourier transformation, pre-processing and features detection, the flourier shape of motion blurred image appears to be a strip, while the defocus blurred image is a cirque. A shape-based identification method is proposed based on the differences of flourier images, in order to automatically identify the two blurred types.The key to image restoration is to estimate PSF. We can calculate PSF by acquiring 3 blurred parameters. To calculate the motion blurred direction, after flourier transformation, noise reduction and edge enhancement, based on the 3*3 differential coefficient kernel, an improving method is proposed in order to improve the precision.With PSF, the thesis restores images with improved winner filter. The method can not only eliminate the noise and enhance the edges, but also eliminate pseudo morph. And the PSNR of improved method is higher than the original method.Although the method of improved winner filter can eliminate noise, the restored images will import new noises. We use larger regularization parameter in the larger region of noise, and smaller regularization parameter in the smaller one. The method is called regularization clearing off noise, with satisfied processing result.We also introduce a regularization method to improve the morbidity according to the morbidity of inversed filter.