We present a number of new piecewise-polynomial kernels for image interpolation. The kernels are constructed by optimizing a measure of interpolation quality based on the magnitude of anisotropic artifacts. The kernel design process is performed symbolically using Mathematica computer algebra system. Experimental evaluation involving 14 image quality assessment methods demonstrates that our results compare favorably with the existing linear interpolators.