Image super-resolution with B-Spline kernels
A novel approach to image super-resolution is described in this paper. By modeling our image acquisition system with a Spline sampling kernel, we are able to retrieve from the samples some statistical information about the observed continuous scene before its acquisition (irradiance light-ﬁeld). This information, called continuous moments, allows to register exactly a set of low-resolution images and to ultimately generate a superresolved image. The novelty of the proposed algorithm resides in its ability to operate entirely on low-resolution images and to enhance the resolution of the entire ﬁeld of view with a relatively low computational complexity. We ran experiments with real acquired images and obtained super-resolved images with a good level of details although some ringing eﬀects are also noticeable.