Interpolation And Fusion Approaches In Image Super-Resolution Reconstruction

Super-resolution reconstruction (restoration) of images is an active research topic in the area of digital image processing. Super-resolution restoration from a single image is the technique which can present more details of the scene when enlarging the image. The purpose of the super-resolution reconstruction from a sequence of images is to reconstruct one or more images of higher resolution from a continuous image sequence of lower resolution by means of information fusion. This thesis investigates both the super-resolution enlargement technique of a single image and the super-resolution reconstruction of image sequences.The thesis presents an interpolation approach based on orientation information measure through the studies of the conventional image interpolation methods and some non-linear interpolation algorithms. The orientation information measure is a tool by which pixels can be classified into one of the following three regions: edge region, smoothness region or texture region. The approach proposed in this thesis takes edge information into account when enlarging the source image. If a pixel is determined on an edge of the image it is interpolated with pixels on the same edge. Otherwise, the pixel is interpolated by bilinear interpolation method. Experimental results show that the approach presented in this thesis is superior to other methods in terms of preserving the edges of the image and reducing the block effect.The super-resolution reconstruction of image sequences is a process which extracts the interested contents from the continuous image sequence. These important contents from two or more successive images are combined to form a new image of high resolution by using a certain fusion rule. The variations of the position and the distance from the interested objects to the camera cause multi-resolution objects in the successive images. Owing to the character of excellent locality and directivity the Contourlet transform is applied to the reconstruction of the multi-resolution objects. Experimental results show that the method achieved satisfactory effect.