This contribution introduces a new method for
lossy image compression which exhibits special advantages when robust compression behavior
in the vicinity of structurally important regions is needed. Comparing to other general
purpose compression techniques, our approach preserves important image characteristics
more precisely while simultaneously proving less susceptible to the introduction of
annoying artifacts. Essentially, the method is based on multilevel polynomial image
interpolation. Aiming at a standard average/detail decomposition of the input image,
polynomial (spline) interpolation on an appropriately chosen (irregular, continuous) grid
enables us to pack more information into the average component and, therefore,
substantially reduces the need for storing additional data pertaining to the detail
component.