A number theoretic transform approach to image rotation in parallel array processors

This paper describes a spectral method for rotating matrix form digital image data in a cartesian-coordinate organized array processor structure. The major elements of the method include: 1) an initial three step number-theoretic-transform (NTT) procedure to implement appropriate column and row data translations, and 2) a final non-spectral translation step within adjacent pixel groups. Digital array image data may be rotated about a selected pivot point through an arbitrary angle within the range ±45°. A comparison of the spectral method execution time with a non-spectral implementation indicates that the new method can provide a performance improvement factor of up to 5 depending upon the magnitude of the rotation angle and pixel word size.