Browse Prior Art Database

MPEG2 Half-Pixel Motion Compensation Using Byte Arithmetic

IP.com Disclosure Number: IPCOM000125084D
Publication Date: 2005-May-18
Document File: 2 page(s) / 31K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method for an algorithm that calculates reference data for half-pixel motion compensation. Benefits include the simultaneous computation of 16 pixels of reference data.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 60% of the total text.

MPEG2 Half-Pixel Motion Compensation Using Byte Arithmetic

Disclosed is a method for an algorithm that calculates reference data for half-pixel motion compensation. Benefits include the simultaneous computation of 16 pixels of reference data.

Background

The following are the current ways to compute reference data for half-pixel motion compensation:

§         Computing reference data using 16 bit arithmetic. Before computing the reference data, 8 bit integer data is loaded into the SIMD register and is extended to 16 bit with zero in 8 most significant bits. The 16 bit data from four locations is summed. The results are shifted to the right by 2 bits. This method requires approximately twice the instructions, because it operates on half the volume of data.

§         Computing reference data using 8 bit arithmetic. This approach consist of two steps. First the intermediate average from the two couples of 8 bit data is calculated using pavgb instructions. Second, the average from the two intermediate results is calculated using pavgb instructions. This method is not precise, and often deviates from exact calculations.

General Description

The disclosed method is an algorithm that calculates reference data for half-pixel motion compensation. The main formula of compensation is:

,

where the data are 8 bit integer values. The result of this equation is also an 8 bit integer value, but the intermediate result is larger. At first sight, it is not possible to calculate reference data, keeping bit-p...