Comparison at high spatial frequencies of two-pass and one-pass geometric transformation algorithms

摘要: Abstract Two-pass image geometric transformation algorithms, in which an is resampled first one dimension, forming intermediate image, then the resulting orthogonal have many computational advantages over traditional, one-pass algorithms. For example, interpolation and anti-aliasing are easier to implement, being 1-dimensional operations; computer memory requirements greatly reduced, with access data external regularized; while pipelined parallel computation simplified. An apparent drawback of two-pass algorithm has tended limit its universal adoption a reported corruption at high spatial frequencies due undersampling, certain cases, necessary image. This experimental study set out resolve question possible by computing mean-square error when sinusoidal grating test rotated, either efficient or traditional algorithm. It was found that method used for major effect on accuracy result, poorer methods accentuating differences between two A totally unexpected fortuitous result that, using near-perfect (e.g., FFT), almost as accurate pleases, rotations up 45°, very close Nyquist (as also algorithm, interpolation). φ > breaks down before limit, but these can be replaced 90° - transposition. Thus, supposed nullified interpolation, least case rotation, bonus greater ease FFT may implemented, case, leading possibility highly faithful practice, aided increasing availability fast DSP microcircuits.