Digital Image Interpolation

The problem of digital image interpolation was posted in 1960s, and from then on, many techniques had been proposed. As it is required in many applica-tions, such as medical imaging and forensics, image demosaicking, remote image geometric rectification, video size conversion, and so on, image interpolation is still a popular research topic in the area of image processing at the present time.In this thesis, we not only give an overview of the recently proposed image interpolation techniques, but also devise a new scheme called trilateral filtering interpolation (TFI), to deal with problems of contour jaggies and blurring arti-facts in conventional image interpolation methods. The proposed method utilizes the smooth varying behaviour of natural image contours, and tries to overcome contour jaggies by suppressing edge convex corner pixels. It first gives a primary estimate of unknown pixels by their four nearest neighbours with proper weights, which are determined by the corresponding edge degree information and spatial distance. Then, in order to give sharper edges, photometric similarities are intro-duced to further refine the weights, based on which, final estimate of unknown pixels obtained. Simulation results show that TFI produces sharper edges than almost all compared interpolation methods, and little jaggies along relatively large object edges. However, in fine texture areas (for example, one pixel objects ap-peared), contour jaggies can not be eliminated effectively. For this reason, we then propose the modified TFI (MTFI), which employs the three nearest neighbours of the unknown pixel rather than four. This MTFI algorithm gives better edge contours in fine texture areas, especially when the edges directed around 45°or 135°.Another feature of the proposed schemes is that they are easily implemented and very efficient. For the tested images, our experiments show that they consume even less time than cubic spline image interpolation. Further more, if we want the algorithms more effective, the photometric similarity refinement could also be canceled, at the expense of more blurring artifacts, thus provide a choice between efficience and performance to users.