The Research Of Image Restoration Method Based On The Nonlocal Theory

Vision is one of the important channels for human to obtain information. In image acquisition, imaging, transmission and processing image, the quality of image will declined due to noise, fuzzy, distortion and so on. The truth is we all need the higher quality image. But we can’t avoid the image degradation, so in recent decades, many scholars have devoted to image restoration. Image restoration is the basis of image processing fields. Image restoration is widely used in many fields, such as computer vision, intelligent transportation, aerospace, medical, etc. So, research image restoration has important meaning and application value.In most cases image restoration method is filtering the image. But the noise of the image and the edge information are all the high frequency information of the image, so how to distinguish the noise from the image edge information is the focus of the filtering model. In recent years, many effective methods for image restoration have been developed, such as wavelet, partial differential equations, neural network, etc.In this paper, the key research is how to apply the nonlocal method for image denoising and deblurring. The main work and innovation is as follows:1. Propose a denoising method based on the nonlocal means filter(NLM) and edge detection.NLM uses highly redundant information of natural images to denoise, and it has a good denoising effect, but the image edge are not preserved well. To overcome this shortcoming, this paper proposes an improved deoising algorithm. The algorithm extracts the lost details of image with edge detection method, then simply adds the lost details back to the denoised image by NLM. We compared the improved denoising algorithm with the NLM and some good classic or the state of the art algorithms experimentally. The results showed the improved algorithm is not only simple, but also restores the original image from noisy image very well while keeping the image edges and details.2. After analyzing several TV models and nonlocal operators. We put forward a new NLTV model. In the process of the Gilboa and Osher’s NLTV model numerical implementation, the problem is to need to know noise variance. The new NLTV in this paper solves the problem. In the experiment, the result showed this new model has higher PSNR than the original NLTV model, and better texture-keeping ability.3. Propose the nonlocal method to decrease the error of uniform linear motion blurring parameter estimation. First, we obtain the PSF of every subwindow in the image, and then, process the PSF by nonlocal means method. Experiments proved that the method has good effect.Finally, this paper gives a summary, and some prospects of further research.