an improved multilevel successive elimination algorithm for fast full- sreach motion estimation tae...

AN IMPROVED MULTILEVEL AN IMPROVED MULTILEVEL SUCCESSIVE ELIMINATION SUCCESSIVE ELIMINATION

ALGORITHM FOR FAST FULL-ALGORITHM FOR FAST FULL-SREACH MOTION ESTIMATIONSREACH MOTION ESTIMATION

Tae Gyoung Ahn, Youg Ho Moon*, and Tae Gyoung Ahn, Youg Ho Moon*, and Jae Ho KimJae Ho Kim

Department of Electronic Engineering Pusan National Department of Electronic Engineering Pusan National University Pusan KoreaUniversity Pusan Korea

OutlineOutline

IntroductionIntroductionMSEAMSEAESEAESEAProposed AlgorithmProposed AlgorithmSimulation ResultSimulation ResultConclusionConclusion

IntroductionIntroduction

Full-search’s heavy computation causes a Full-search’s heavy computation causes a problem for real-time Application problem for real-time Application

Fast full-search algorithm been proposed Fast full-search algorithm been proposed to provide the same accuracy with small to provide the same accuracy with small computationcomputation

Successive elimination algorithm ( SEA )Successive elimination algorithm ( SEA )- Computation saved by omitting the SAD - Computation saved by omitting the SAD

calculation for the invalid candidate block calculation for the invalid candidate block

Introduction Introduction

N-1,N-1N-1,N-1

SAD SAD (m, n)(m, n) = = ∑ ∑ |∑ ∑ |ff( ( i, j, ti, j, t ) - ) - ff( ( i - m, j - n, ti - m, j - n, t -1 )| -1 )|

i=0,j=0 i=0,j=0 （（ 11））for a N*N candidate block for a N*N candidate block

- - ff( ( i, j, ti, j, t ) and ) and ff( ( i, j, ti, j, t -1 ) represent an intensity of pixel -1 ) represent an intensity of pixel ( i, j ) in the current frame t and the previous ( i, j ) in the current frame t and the previous frame t-1 frame t-1

MSEAMSEA

MSEAMSEA

Subblock SAD at level Subblock SAD at level ll is defined as is defined as ：：

（（ 22 ））

It can be driven that :It can be driven that :

（（ 33 ））

MSEAMSEA

Invalid candidate block is determined by cInvalid candidate block is determined by comparing the previously obtained minimuomparing the previously obtained minimum SAD ( SAD m SAD ( SAD minmin ) with SSAD ) with SSADl l at each leveat each leve

ll

ESEAESEA

Many invalid candidate blocks are eliminated by Many invalid candidate blocks are eliminated by MSEA. But the remaining blocks have to obtained MSEA. But the remaining blocks have to obtained SAD SAD (m, n)(m, n)

( 4 )( 4 )

d( d( i, ji, j ) = ) = ff( ( i, j, ti, j, t ) - ) - ff( ( i, j, ti, j, t -1 ) -1 )

ESEA remove the overhead by lookup tableESEA remove the overhead by lookup table

Proposed algorithmProposed algorithm

An improved MSEA reducing the An improved MSEA reducing the computations required to judge invalid computations required to judge invalid candidate blockcandidate block

Rewrite decision condition Eq. ( 3 )Rewrite decision condition Eq. ( 3 )

Proposed algorithmProposed algorithm

DDll( u, v ) ( u, v ) the difference between the current and the difference between the current and candidate block at level candidate block at level ll

DDll( u, v ) = R( u, v ) = Rll( u, v ) – M( u, v ) – Mll( u, v ) ( u, v ) 0 <= u,v < 20 <= u,v < 2ll –1 ( 5 ) –1 ( 5 )

The difference between RThe difference between R00 and M and M0 0 (m, n)(m, n)

can can be described as :be described as :

( 6 )( 6 )

Proposed algorithmProposed algorithm

According to Eq. ( 2 ) and Eq. ( 6 ) According to Eq. ( 2 ) and Eq. ( 6 )

SSADSSADll - | R - | R00 - M - M0 0 (m, n)(m, n) | | ( for ( for (( R R00 - M - M0 0

(m, n)(m, n) )) < 0 ) < 0 )

Proposed algorithmProposed algorithm

The same manner for The same manner for (( R R00 - M - M0 0 (m, n)(m, n) )) >=>= 0 0

then we can obtain follows : then we can obtain follows :

Proposed algorithmProposed algorithm

We can rewrite Eq.( 3 ) as follows :We can rewrite Eq.( 3 ) as follows :

Define ConDDefine ConDll = ( SSAD = ( SSADll - | - | RR00 - M - M0 0 (m, n)(m, n) | ) / 2| ) / 2

Cond = ( SSACond = ( SSA( m ,n)( m ,n) - | - | RR00 - M - M0 0 (m, n)(m, n) | ) / 2| ) / 2

Proposed algorithmProposed algorithm

New decision condition : New decision condition :

The ConDThe ConDll is partial sum of D is partial sum of D ll( u,v )( u,v )

Save the computationSave the computation

Simulation resultsSimulation results

Simulation resultsSimulation results

ConclusionConclusion

Without degradation accuracy, propose alWithout degradation accuracy, propose algorithm calculation SAD and SSAD using tgorithm calculation SAD and SSAD using the already obtained | he already obtained | RR00 - M - M0 0

(m, n)(m, n) | | New decision condition reduces complexitNew decision condition reduces complexit

y for invalid candidate blocks y for invalid candidate blocks Improve MSEAImprove MSEA