Low-Drift Fixed-Point 8X8 IDCT Approximationwith 8-Bit Transform Factors

We describe an efficient algorithm for computing the inverse discrete cosine transform (IDCT) for image and video coding applications. This algorithm was derived by converting an 8-point IDCT factorization of C. Loeffler, A. Ligtenberg, and G. Moschytz into a scaled form, leaving 8 multiplications by irrational factors inside the transform. The key advantage of such a modification is that these factors can be sufficiently accurately represented by 8-bit integer values, resulting in a very small dynamic range of variables inside the transform. Our scaled ID transform can be implemented either by using 8 multiplications, 26 additions and 6 shifts or (in a multiplier-less fashion) by using only 44 additions and 18 shifts. This implementation fully complies with the new MPEG IDCT precision standard (ISO/IEC 23002-1, replacement of former IEEE 1180 specification), and shows remarkably low drift in decoding of H.263, MPEG-2, and MPEG-4 bitstreams produced by reference software encoders (employing 64-bit floating-point DCT and IDCT implementations).