Accuracy requirements standardization for 8x8 IDCT
MPEG doc#: N7326
Date: July 2005
Author: Gary Sullivan
A number of visual coding related specifications (e.g., MPEG-1 Part 2, MPEG-2 Part 2 / H.262, MPEG-4 Part 2, JPEG-1, H.261, and H.263) include a requirement for decoders to implement an integer-output approximation of the 8x8 inverse discrete cosine transform (IDCT) function.
These relevant standards require decoders to use an approximation that is within some degree of accuracy relative to the integer values that would be obtained from the ideal real-valued function definition of the IDCT, followed by rounding to the nearest integer. The approximation requirements in the relevant video coding standards are largely similar, although differing in details.
The development of a single MPEG standard containing the technical content for specification of the various parameters of IDCT accuracy requirements is necessary to establish these requirements in an unambiguous fashion. The development of such a standard further allows conformance requirements for various other standards to be established (either partly or entirely) by reference to a single specification.