Browse Prior Art Database

Motion estimator implementing stepping scheme for maximum use of local memory

IP.com Disclosure Number: IPCOM000032186D
Publication Date: 2004-Oct-26
Document File: 2 page(s) / 31K

Publishing Venue

The IP.com Prior Art Database

Abstract

ID298922

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 74% of the total text.

Page 1 of 2

Motion estimator implementing stepping scheme for maximum use of local memory

Field of invention:

This invention relates to real time digital video MPEG compression techniques. Objective for these techniques is a good resulting picture quality at low bit rates. MPEG video compression also uses temporal redundancy reduction based on motion estimation, which is found to be one of the key elements influencing the picture quality.

Especially MPEG video compression needs a large amount of memory for de-interlacing and predictive encoding. This memory is most effectively built externally for current CMOS processes. This is, why current platform approaches combine scalable blocks, a predefined bus structure and large external memories. Each component that needs to access the memory has to share the bandwidth with others; hence bandwidth to external memory is a key metric in the design of such components.

Motion estimation with the state of the art block matching algorithms is one of the most bandwidth-consuming building blocks of an MPEG encoder. Especially when moving to higher resolutions, internal memories are required which should be optimally used.

State of the art:

Current systems process blocks of pixels in scan order or use dedicated ordering schemes to reuse locality in the memory accesses (e.g. patent US 6,163,576).

Description of the invention:

In today's compression standards, such as MPEG-1, MPEG-2, MPEG-4 as well as H.264 macro blocks of 16 by 16 pixels are building a unit for diverse operations and decisions. Some macro blocks are depending on results of others due to predictive coding. These results could either be pixel data and/or motion information. In the mentioned compression standards macro blocks (or in H.264 with MBAFF macro block pairs) depend on results of neighboring macro blocks on the left, upper left, upper as well as upper right position. This means that the shaded macro blocks as shown in Figure 1 are independent from each other and can be independently processed.

Figure 1: independent blocks

Page 2 of 2

Regarding only motion estimation, this means, that there is a large overlap of search areas of all independent macro blocks, which implies data locality (see Figure 2 with e.g. a fixed...