Enhanced Enhanced Hexagonal Search Hexagonal Search

for Fast Block for Fast Block Motion EstimationMotion Estimation

AuthorsAuthors ：： Ce Zhu, Xiao Lin, Ce Zhu, Xiao Lin, Lappui Chau, and Lai-Man Lappui Chau, and Lai-Man PoPoIEEE TRANSACTIONS ON CIRCUITS AND SYIEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, OCTOBESTEMS FOR VIDEO TECHNOLOGY, OCTOBER 2004R 2004

OutlineOutline

IntroductionIntroduction HEXBS ( Hexagon-Based Search )HEXBS ( Hexagon-Based Search ) Predictive HEXBSPredictive HEXBS Fast Hexagonal Inner Search Fast Hexagonal Inner Search 6-Side-Based Fast Inner Search 6-Side-Based Fast Inner Search Enhanced HEXBS Algorithm Enhanced HEXBS Algorithm Experimental Results And Analysis Experimental Results And Analysis Conclusion Conclusion

IntroductionIntroduction

FastFast block motion estimation process block motion estimation process ：： STEP 1STEP 1：： low-resolution coarse searchlow-resolution coarse search

→ → To identify a small area To identify a small area where the where the best motion vector is expected best motion vector is expected

to lie to lie STEP 2STEP 2：： fine-resolution inner search fine-resolution inner search

→ → To select the best motion vector in To select the best motion vector in

the located small region the located small region

Most motion estimation algorithms attempt to Most motion estimation algorithms attempt to speed up the coarse search without considering speed up the coarse search without considering accelerating the inner search accelerating the inner search

Enhanced hexagonal search algorithm is propoEnhanced hexagonal search algorithm is proposed to improve the performancesed to improve the performance ：：

(1) Reducing number of search points (1) Reducing number of search points

(2) Decrease the distortion (2) Decrease the distortion

IntroductionIntroduction

The two-dimensional logarithmic searchThe two-dimensional logarithmic search Three-step search (TSS)Three-step search (TSS) New three-step search (NTSS) New three-step search (NTSS) Four-step search (4SS) Four-step search (4SS) Block-based gradient descent search (BBGDS) Block-based gradient descent search (BBGDS) Simple and efficient search (SES) Simple and efficient search (SES) Diamond search (DS)Diamond search (DS) The hexagonal search The hexagonal search Hexagon-based search (HEXBS) Hexagon-based search (HEXBS) Enhanced hexagon-based search Enhanced hexagon-based search

IntroductionIntroduction

HEXBSHEXBS

Inner points in the hexagonal search pattern

HEXBSHEXBS

Using the shrunk hexagonal pattern covering the pUsing the shrunk hexagonal pattern covering the points 2, 4, 5, and 7oints 2, 4, 5, and 7

A gradient schemeA gradient scheme ：： ExEx ：： Points 1 and 3 if point 2 wins in the last step Points 1 and 3 if point 2 wins in the last step of the HEXBS algorithm of the HEXBS algorithm

The inner search for the The inner search for the HEXBSHEXBS

The flowchart of HEXBSThe flowchart of HEXBS

NHEXBS (mx , my) = 7 + 3n + 4 NHEXBS (mx , my) = 7 + 3n + 4 where n is the number of times of lowhere n is the number of times of low-resolution coarse searchw-resolution coarse search

Predictive HEXBS Predictive HEXBS

The error distortion function has monotonic The error distortion function has monotonic characteristic in a localized search area characteristic in a localized search area

The motion vector of the current block is The motion vector of the current block is highly correlated to those of its neighboring highly correlated to those of its neighboring blocks. blocks.

The motion information of neighboring blocks The motion information of neighboring blocks can be utilized for prediction of a good starting can be utilized for prediction of a good starting point point

Consider the upper and the left neighboring Consider the upper and the left neighboring blocks blocks

Finding a good starting point using the Finding a good starting point using the neighboring motion vectors neighboring motion vectors

Normally finds better motion vectors than the Normally finds better motion vectors than the original HEXBS schemeoriginal HEXBS scheme

Predictive HEXBS Predictive HEXBS

Not a full inner search , only check a portion of Not a full inner search , only check a portion of the inner search pointsthe inner search points

Strong correlation exists between the inner Strong correlation exists between the inner search points search points

Based on the monotonic distortion characteristic Based on the monotonic distortion characteristic in the localized area around the global minimumin the localized area around the global minimum

Fast Hexagonal Inner Fast Hexagonal Inner Search Search

Group the search points in the six sides of the Group the search points in the six sides of the hexagon hexagon

Define a group distortion by summing the Define a group distortion by summing the distortions of all the points within the group distortions of all the points within the group

We focus the inner search just in the region We focus the inner search just in the region near to the group with the smallest group near to the group with the smallest group distortiondistortion

For different groups (sides) in different For different groups (sides) in different locations, we have different number of inner locations, we have different number of inner search points search points

6-Side-Based Fast Inner 6-Side-Based Fast Inner SearchSearch

HEXBS incorporate the 6-side-based fast inneHEXBS incorporate the 6-side-based fast inner search scheme r search scheme

Moreover , incorporate the Predictive HEXBSMoreover , incorporate the Predictive HEXBS The reduction of number of search points for tThe reduction of number of search points for t

he enhanced HEXBS algorithm he enhanced HEXBS algorithm ：： (1) The prediction for a good starting point(1) The prediction for a good starting point

using the predictive HEXBS, using the predictive HEXBS,

(2) The fast inner search. (2) The fast inner search.

Enhanced HEXBS Enhanced HEXBS AlgorithmAlgorithm

Experimental Results And Experimental Results And Analysis Analysis

TABLE III TABLE III

100%)N

N-N(%SIR

1

12

100%)MSE

MSE-MSE(%D

1

12MSE

, where Ni is the number of search points used in the Method i

,where MSEi is the distortions for Methods i

Conclusion Conclusion

Enhanced HEXBS speeds up the motion Enhanced HEXBS speeds up the motion estimation and decreases distortionsestimation and decreases distortions

Only part of the inner points will be evaluatedOnly part of the inner points will be evaluated Enhanced HEXBS algorithm Enhanced HEXBS algorithm

outperforms the original HEXBS outperforms the original HEXBS