Based On The Fuzzy Image Of The Frequency-domain Characteristics Of Digital Image Blind Restoration Algorithm

With the rapid development of computer science and technology, digital images processing technology has been used in many fields such as astronomy, traffic and medical imaging, so in most realistic circumstance it demands corresponding more better image processing technologies. In reality, people always expect images with good quality。But with many objective limits such as the performance of imaging devices, the circumstance of imaging and the human factors etc, the images we get are always not so desirable in most cases. However, it's not so acceptable to get images with good quality by improving the performance of imaging devices because it's a greater cost. So in such a case ,when people want images with good quality, some good post-processing algorithms-image restoration techniques are expected to improve image quality.There are two main blur types: motion blur and defocus blur. And the estimation of point spread function in blurred image is the key of whole image restoration. Approximately, we can consider a blurred image as the convolution of its original clear image and its point spread function. And image restoration is the opposite course of convolution , so-called deconvolution.Considering common problems in image restoration field, the article proposes a method to restore blurred images, based on characteristics of image in frequency domain. The first one is about motion blur image. First, motion direction is estimated by radon transform and during the course we use binary characteristics to improve the accuracy of estimation. After knowing the motion direction, we can get motion length by fourier transform and cepstrum and then we can use the two parameters to restore blurred image and get restored images with good quality. The second one is about defocused image. We use characteristics of cepstrum in frequency domain and its corresponding correlation to estimate the only parameter of point spread function- blur radius, then restore the blurred image with classical methods and get restored images with good quality. Finally ,a new method to evaluate the clearness of restored images is proposed ,based on modulous gradient of un-decimated wavelet decomposition.The experiments results show that proposed algorithms can estimate PSF values accurately and have a good restoration result. The result shows that the evaluation principle of restored images ,which coincides with the visual system of people, is reasonable and effective.