Browse Prior Art Database

Motion vector grouping for search position reduction in hierarchical video motion estimators Disclosure Number: IPCOM000032187D
Publication Date: 2004-Oct-26
Document File: 3 page(s) / 20K

Publishing Venue

The Prior Art Database



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 51% of the total text.

Page 1 of 3

Motion vector grouping for search position reduction in hierarchical video motion estimators

Field of invention:

Motion estimation is known as being the most computation power consuming part in real time digital video compression techniques. On the other hand, temporal redundancy reduction based on motion estimation is one of the key elements influencing the picture quality. That is, the main focus in video compression has to be putted on the selection process of motion vectors.

A lot of investigations have been made in order to reduce the number of motion vector candidates that have to be refined in hierarchical video motion estimators. Hierarchical video motion estimation in current standards consists of three steps: search on full-pel, half -pel and finally quarter-pel resolution.

The following invention describes an efficient way to reduce dramatically the amount of motion vector candidates while conserving high coding efficiency.

Description of the invention:

In current compression standards, such as MPEG-4, H.264, as well as Windows Media 9 the amount of possible macroblock partitions is high. A macro block may be partitioned using up to 7 different partitioning modes (e.g. up to 4x4 block partitions in H.264).

If full-pel search operation is execute d for every possible block partition, it will result in up to 41 different full-pel search vectors (e.g. H.264) each block having its own independent motion vector.

In hierarchical motion search (full-pel, half-pel then quarter-pel) only some few accurate candidates should be derived from upper hierarchical steps in order to reduce search positions. Taking all possible candidates for instance will result in 1681 (41*41) candidates for one macroblock in H.264. Motion estimators can hardly handle this amount of motion vector candidates.

The proposed search strategy reduces the number of sub-pel search candidates by combining identical or neighboring full-pel vectors according to different criteria before using them as search center for a further refinement step. The criteria for refinement candidate selection are:

· Macroblock partition size (area)
· Sum of absolute differences between current and predicted block (SAD)
· Number of neighboring motion vectors


Fast motion estimation, hierarchical search, sub-pel motion vector selection, motion vector grouping,

Detailed description of the invention:

Some parameters are necessary for motion vector grouping:

RANGE_X width of window that contains a vector group in a close neighborhood RANGE_Y height of window t hat contains a vector group in a close neighborhood NUMB_OF_VECTORS_IN_GROUP number of motion vectors that are in one group

Page 2 of 3

NOMB_OF_WINNER_GROUPS number of groups or number of vectors to be refined in the next hierarchical search step

AREA area of the block that provides a motion vector (e.g. 256 for a 16x16 block)

SAD sum of absolute differences between current and predicted block

WEIGHT weighting criteri...