The Application Of Bi-cubic COONS Surface In Sequence Images Super-resolution

High resolution images are needed in more and more electronic fields. But most digital digitizers can only record images at a resolution lower than desirable. Due to the limitation of hardware, higher expense will be paid out if hardware is improved on. So, effective method for improving images resolution has become the aim which people are pursuing. Super-resolution reconstruction is the most important and basal technique in digital image processing. It can help us to solve this question. Super-resolution from a single low-resolution, and possibly blurred, image is known to be highly ill-posed. However when a sequence of low-resolution frames is available, such as those obtained by a video camera, the problem becomes more manageable. It is evident that the 3-D spatio-temporal sampling grid contains more information than 2-D still-frame sampling grid. Inter-frame super-resolution methods exploit this additional information, contained in multiple frames, to reconstruct a high-resolution still image or a sequence of high-resolution images.The key to the super-resolution image reconstruction is the accurate knowledge of the sub-pixel translational motion of the neighbor frames and the accurate estimation of the missing frames (a frame whose translational shifts are similar to another frame or a frame which does not exist corresponding to a desired grid pattern). And getting the accurate sub-pixel motion information is a main task for computer vision and image registration. So super-resolution image reconstruction consists of image registration and image reconstruction. In this thesis, we introduce bi-cubic Coons surface fitting to the field of image registration. We use a piecewise bi-cubic Coons surface to simulate the local image model to approximate the original image. Then we registered the sequence images through the local image model. When we get the more accurate sub-pixel motion information among the related frame, we use an iterative method to reconstruct the high resolution images through interpolation. The numeric experiments and the result higher resolution images prove the validity of this method.