Order Statistic Filters for Image Interpolation

Image interpolation techniques are commonly used for converting low-resolution images into high-resolution images. Linear interpolation methods usually blur image details. In this paper, an order statistic filtering algorithm which preserves fine object details is proposed for image resolution up-conversion. Pixels in the filter aperture are first ordered according to their spatial distances to the new pixel to be interpolated. Pixels with the same spatial distance to the new pixel are further ordered according to their intensity deviations to the central pixel in the aperture. The optimal filter coefficients are obtained by statistical training on a dataset which is composed of the original high-resolution images and the down-sampled versions of the original images.