Biomedical image interpolation based on multi-resolution transformations

We present a novel feature-based image interpolation approach. Two continuous maps for the image domain are constructed via an L-gradient flow based on their multi-resolution representations so that the features of the given images are matched at multiple scales. The flow equation is efficiently solved using a finite element method in the bicubic B-spline vector-valued function space. The interpolated images are then obtained from the domain maps at any sampling rate. Experimental results show that our interpolation approach is effective, capable of capturing image features from large to small. It yields continuously and uniformly deformed inbetween images.