Intel® Integrated Performance Primitives Developer Guide and Reference

ID 790148
Date 11/07/2023
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents

Discrete Cosine Transforms

Discrete Cosine Transform (DCT) of a real 2D image yields output results that are also real, which eliminates the need to use packed format for storing the transformed data. However, forward and inverse DCT functions ippiDCTFwd and ippiDCTInv need different context data structures to be initialized and filled in prior to their use. Consequently, the required workspace buffer size is different for these functions. In case of using an external buffer, its size must be determined by previously calling the respective support function. DCT functions that use context structures implement the modified computation algorithm proposed in [Rao90].

The DCT functions ippiDCT8x8Fwd and ippiDCT8x8Inv working on a fixed 8x8 image buffer need no context data or external workspace buffers. Functions ippiDCT8x8Inv meet IEEE-1180 standard requirements (see [IEEE]).

Intel IPP Discrete Cosine Transform functions working on a fixed 8x8 image buffer use Feig and Winograd algorithm ([Feig92]) modified for taking advantage of SIMD instructions. For details on algorithms used in DCT transforms and for more references, see [AP922].

Product and Performance Information

Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.

Notice revision #20201201