Sub-pixel registration based super resolution

Nowadays, depending on the developments and expanding application areas of imaging systems, demand for more detailed and high resolution (HR) images increased. However, it is not always possible to get HR images because of the cost of imaging system and storage. Because of this, it is needed to use some image processing algorithms to obtain HR images from low resolution (LR) images. Super resolution (SR) methods are utilized to solve this problem. SR algorithms can be basically inspected under two main titles. First of these is to obtain HR image from only one reference LR image. This process is applied by using an interpolation method. The reference image is expanded to the size demanded. Then, new pixels are filled depending on the interpolation method. Second is using more than one LR images. By this method, the reference images of a view from one or more camera at the same time are used to get HR images. The principal of this method is to use shift information between reference images. This information can be found by various registration methods. The performance of the registration method directly affects the achievement of SR. The accuracy of SR using more than one LR images is higher than the results from the SR methods using only one LR image. In this thesis, first, the size of the reference image is upsampled and from the upsampled image, new images which are same sized with reference image are created. These virtual LR images are registered using shift information and HR image is obtained. It is analyzed that our proposed method gives better results than the other registration methods.