Arianne T. Hinds

Efficient Fixed-Point Approximations of the 8x8 Inverse Discrete Cosine Transform

This paper describes fixed-point design methodologies and several resulting implementations of the Inverse Discrete Cosine Transform (IDCT) contributed by the authors to MPEGs work on defining the new 8x8 fixed point IDCT standard - ISO/IEC 23002-2. The algorithm currently specified in the Final Committee Draft (FCD) of this standard is also described herein.

Low Complexity Fixed-Point Approximation of Inverse Discrete Cosine Transform

This paper presents an efficient algorithm for computing the Inverse discrete cosine transform (IDCT) for image and video coding applications. This algorithm was submitted in response to MPEGs call for proposals for ISO/IEC 23002-2 (fixed-point 8times8 IDCT and DCT) standard, and was subsequently adopted in the Working Draft 1 of this standard. Our proposed algorithm is a multiplication-free implementation. It is based on a modification of Arai, Agui, and Nakajimas (AAN) factorization, and requires only 42 addition and 16 shift operations per scaled ID transform. Each register in our scaled ID transform requires at most 22 bits. This implementation complies with the MPEG IDCT precision specification ISO/IEC 23002-1.

Multicriterial Optimization Approach to Eliminating Multiplications

Widely used image compression algorithms, such as discrete cosine transform (DCT), can be described as applying a linear operator to a vector of samples. This is computationally quite complex and a number of algorithms have been developed to minimize the number of necessary operations, particularly multiplies. This paper describes a method to find approximations of the remaining multiplication constants in these algorithms, such that the approximation error is minimized for each substitution of multiplications by shifts and adds

Enhanced parallel processing in wide registers

Wide computer registers offer opportunities to exploit parallel processing. Instead of using hardware assists to partition a register into independent non-interacting fields, the multiple data elements can borrow and carry from elements to the left, and yet be accurately separated. Algorithms can be designed so that they execute within the allocated precision. Their floating point or irrational constants (e.g., cosines) are converted into integer numerators with floating point denominators. The denominators are then merged into scaling terms. To control the dynamic range and thus require less bits of precision per element, shift rights can be used. The effect of the average truncation errors is analyzed and a technique shown to minimize this average error.

Drift Analysis for Integer IDCT

This paper analyzes the drift phenomenon that occurs between video encoders and decoders that employ different implementations of the Inverse Discrete Cosine Transform (IDCT). Our methodology utilizes MPEG-2, MPEG-4 Part 2, and H.263 encoders and decoders to measure drift occurring at low QP values for CIF resolution video sequences. Our analysis is conducted as part of the effort to define specific implementations for the emerging ISO/IEC 23002-2 Fixed-Point 8x8 IDCT and DCT standard. Various IDCT implementations submitted as proposals for the new standard are used to analyze drift. Each of these implementations complies with both the IEEE Standard 1180 and the new MPEG IDCT precision specification ISO/IEC 23002-1. Reference implementations of the IDCT/DCT, and implementations from well-known video encoders/decoders are also employed. Our results indicate that drift is eliminated entirely only when the implementations of the IDCT in both the encoder and decoder match exactly. In this case, the precision of the IDCT has no influence on drift. In cases where the implementations are not identical, then the use of a highly precise IDCT in the decoder will reduce drift in the reconstructed video sequence only to the extent that the IDCT used in the encoder is also precise.

On the coding of interlace scanned content in HEVC

High Efficiency Video Coding is the latest in the series of video coding standards developed either by MPEG, or VCEG, or jointly through a collaboration of the two committees. The first version of HEVC was completed in January 2013, but was developed without specific requirements for the compression of interlace video content. Rather, the requirements for the initial version of HEVC targeted the reduction, by 50%, of the bitrate required to delivery progressive video signals at the same, or nearly the same, visual quality as achieved by current state-of-the-art video codecs. Despite the lack of formal requirements for the support of interlace scanned content, this first version of HEVC nevertheless supports interlace video formats but achieves this support in a nominal manner, without the use of specific coding tools. Interlace formats, however, continue to be the primary format used by broadcasters for the capture and delivery of video being most recently used exclusively to capture and broadcast the 2012 Summer Olympics for the entire world. This paper explores the continued importance and relevance of interlace formats for next generation video coding standards, including HEVC. The in-progress experiments and results of a formal study of HEVC for the coding of interlace content are presented.

CIELAB to CMYK color conversion in the transform domain for JPEG compressed images

A new technique is described for color conversions of JPEG images. For each input block of each component, the conversion for the 63 AC coefficients is processed in the transform domain instead of the spatial domain. Only the DC coefficients for each input block of the color components are transformed to the spatial domain and then processed through the traditional lookup table to create color-converted output DC coefficients for each block. Given each converted DC value for each block, the remaining 63 AC coefficients are then converted directly in the transform domain via scaling functions that are accessed via a table as a function of only the DC term. For n-dimensional color space to m-dimensional color space conversion, n component blocks create m component blocks. An IDCT can then be applied to the m component blocks to create spatial domain data or these output blocks can be quantized and entropy encoded to create JPEG compressed data in the m-dimensional color space.