Research For Image Super-resolution Reconstruction Algorithm Based On Interpolation

The super-resolution of images and videos is a hot issue under research, it has important application in high-definition television, image coding, face recognition, medical image and safety surveillance etc. The image resolution represents the detail of the image, the higher the resolution, and the more details. The classical image super-resolution algorithm firstly interpolates the low resolution image, and then establishes the corresponding mathematical model for restoration and reconstruction of the image. The performance of image interpolation algorithm will directly affect the quality of image super-resolution reconstruction results; a well performance interpolation algorithm can greatly reduce the workload of restoration and reconstruction. However, in the applications of the image super-resolution, due to computational resource constraints, complex methods cannot be used, even the steps of image restoration and reconstruction are omitted. So a more stable and effective image interpolation method is needed. This paper presents an edge preserving filtering algorithm for image interpolation.The main purpose of the super-resolution image is de-noising, de-blurring and up-sampling, so as to increase the resolution of image or video.The super-resolution reconstruction of the noisy image and the blurred image is being mainly introduced in this thesis. In order to deeply study the super-resolution reconstruction algorithms of noisy image, this thesis mainly analyze the common image and the mixed noise image de-noising algorithms, and study the super-resolution image algorithms based on the wiener filter theory and the non-local mean decomposition and reconstruction algorithms. Firstly, every pixel of input noised image should be checked whether it’s the pepper noised point or not. If it’s true, this pixel will be filtered by the median filtering. Secondly, the image that we got from the upper step filtered by Wiener filtering and non-local mean filtering to get the basic image layer. The image layer that contain noise and texture information can be got from the de-noised image minus the basic image layer, which we call it detailed layer. Meanwhile, the outline information of image mainly store in the basic layer. So we filter the detail layer again to de-noising the noise. Finally, both of the detailed layer and the basic layer should be edge-keeping interpolated before we fuse the detailed layer and the basic layer to get the output image.Research on super-resolution reconstruction algorithm of fuzzy image, mainly analyzes the amplification restoration of motion blur and defocus blurred image. However that the key of fuzzy image correctly recover is the point spread function. At first, the paper introduces the fuzzy pattern recognition, fuzzy parameter estimation, and then analyzes the method of the fuzzy image restoration. The entire process is fuzzy type judgment based on Hough transform, and then using wiener filtering and the estimated point spread function restoration image. Finally, high resolution image can be obtained by magnification image using edge preserving to obtain high resolution image.For the low-resolution image that polluted by the noise and blur, the image super-resolution based on the kurtosis can get the better effect, but slower. Thus, we improve it by using the DOG theory in this paper for speeding the progress.