Research On Image Super-resolution Reconstruction

High resolution means that pixel density within an image is high, and therefore a high resolution image can offer more details. It provides a good basis for image post-processing. However, confined by the imaging devices, illumination and so on, images with a low resolution are obtained usually. So, how to improve the quality of a image effectively becomes a very crucial and important tasks in image processing.Super-resolution image reconstruction is one of the principal means to improve image resolution currently. It reconstructs a high resolution image from a set of low resolution images of a scene. Precise alignment of the input images is a crucial step of such algorithms. Low resolution images are often under-sampled and have aliasing artifacts, so the performance of standard registration algorithms based on Fourier transform decreases. Moreover, conventional rotation estimation methods based on two-dimensional correlation approach is computationally heavy and thus practically difficult.In view of the above problems existed in multi-frames super-resolution, this paper carried out a detailed study and experimental analysis, research works can be described as follows:First, an introduction of the development status both at home and abroad was given. Image registration, interpolation and image restoration, the three stages included by most super-resolution algorithms were summarized and the most intuitive method non-uniform interpolation approach was discussed.Second, image registration method based on frequency domain was analyzed and the pitfalls existed in the method of converting two-dimensional correlation to one-dimensional correlation was discussed. A weighted approach is presented to improve the registration results.Third, based on the weighted approach, a global overview of the super-resolution algorithm is given. Standard images were used in simulation to verify the effectiveness of the algorithm. At the same time, the experimental results on real image showed the practicability of this algorithm.Finally, the research work of this paper is summarized and some constructive approaches for post-work were given.