Accurate image rotation using DCT transformation

Image rotation is an important part of image processing for its wide range of uses. High precision and stability are demanded in many fields, and subtle errors may cause completely different judgment. With conventional three interpolation approaches: nearest, bilinear and bicubic, they cannot reach accuracy requirement. Using Hermite method appears to be accurate in most cases, but it's accuracy is extremely low in a small range. In this paper, we propose a novel image rotation method. This method takes out an n × n block around the point which is the nearest integer point around the unknown point and performs 2-dimensions DCT transformation in the block to obtain the frequency domain information, then interpolate the unknown pixel by performing its 2-dimensions IDCT in the n × n block. Our method can be not only more accurate than the conventional interpolation but more steady than Hermite method.