'
EVALUATION AND PRACTICAL ISSUES OF SUBPIXEL IMAGE REGISTRATION USING PHASE CORRELATION METHODS
$
Fabrice Humblot1,2 , Bertrand Collin1 and Ali Mohammad-Djafari2 1
DGA/DCE/CTA/DT/GIP
2
Centre Technique d’Arcueil
LSS (CNRS-SUPELEC-UPS) ´ ´ Ecole Sup´erieure d’Electricit´ e
94114 Arcueil, FRANCE.
91192 Gif-sur-Yvette, FRANCE.
[email protected]
[email protected] [email protected]
&
1st February, 2005 – PSIP 2005 – Toulouse 1
%
$
'
Contents • Principle of the Phase Correlation Method • Extension to Subpixel Registration • Extension Using Image Gradients • Registration Equation • Numerical Evaluation • Evaluation with Real Unknown Data • Conclusions &
2
%
$
'
Principle of the Phase Correlation Method [Kuglin & Hines, 1975] I2 (x, y) = I1 (x − x0 , y − y0 ) Iˆ2 (u, v) = Iˆ1 (u, v)e−j(u×x0
+ v×y0 )
Iˆ2 (u, v)Iˆ1∗ (u, v) −j(u×x0 = e |Iˆ2 (u, v)Iˆ∗ (u, v)|
+ v×y0 )
1
δ(x − x0 , y − y0 ) &
3
%
$
'
Extension to Subpixel Registration [Foroosh, Zerubia & Berthod, 2002] Iˆs1 (u, v) =
1 MN
M −1 N −1 X X
Iˆ
u+2πm v+2πn , N M
m=0 n=0
Iˆs2 (u, v) =
1 MN
M −1 N −1 X X
Iˆ
u+2πm v+2πn , N M
m=0 n=0
C(x, y) = IFT &
4
e−j(
∗ (u,v) Iˆs2 (u,v)Iˆs1 ∗ (u,v)| |Iˆs2 (u,v)Iˆs1
u+2πm x0 , v+2πn y0 M N
)
%
$
'
Extension to Subpixel Registration (2) C(x, y) ≈
∆x
=
x0 M
sin(π(M x−x0 )) sin(π(N y−y0 )) π(M x−x0 ) π(N y−y0 )
=
C(xs ,yp )xs ±C(xp ,yp )xp C(xs ,yp )±C(xp ,yp )
xs = x p ± 1 ys = yp ± 1
&
5
%
$
'
Extension Using Image Gradients [Argyriou & Vlachos, 2003]
Gh (x, y) = I(x + 1, y) − I(x − 1, y)
Gv (x, y) = I(x, y + 1) − I(x, y − 1)
G = Gh + j × G v &
6
%
$
'
Registration Equation Bilinear Approximation dx
x y
1
2
3
4
5
6
7
&
8
dy
9
1
2
3
4
5
6
x = X + d (X, Y ) ∈ Z2 0 x y0 = Y + dy (dx , dy ) ∈] − 1; 1[2 7
%
'
Registration Equation Bilinear Approximation (2)
$
I1 (x, y) = I0 (x − X, y − Y ) avec (X, Y ) ∈ Z2
I2 (x, y) =
&
(1 − |dx |)(1 − |dy |)I1 (x, y) + |dx |(1 − |dy |)I1 (x − sign(dx ), y) + (1 − |dx |)|dy |I1 (x, y − sign(dy )) + |dx ||dy |I1 (x − sign(dx ), y − sign(dy )) 8
%
$
'
Numerical Evaluation Accurate Case of an Analytical Function
&
xˆ = 1.340 0 yˆ0 = 1.847
x = 1.350 0 y0 = 1.850 9
%
$
'
s[i, j] =
&
Z
i+0.5 i−0.5
Z
j+0.5
h0 (u, v) du dv j−0.5
10
%
'
$
&
%
Error on dx Estimations
11
$
'
Neglicence of Integration Feature
&
12
%
'
Simulated Movement Using a Real Image
&
x = 10.30 0 y0 = 12.60 13
$
%
'
Simulated Movement Using a Real Image (2)
x = 10.30 0 y0 = 12.60
&
xˆ = 10.16 0 yˆ0 = 12.69 14
xˆ = 10.17 0 yˆ0 = 12.70
$
%
'
$
&
%
Evaluation with Real Unknown Movement
15
'
Evaluation with Real Unknown Movement (2)
x =? 0 y0 = ?
&
xˆ = −0.07 0 yˆ0 = −0.12 16
xˆ = 9.51 0 yˆ0 = 2.48
$
%
$
'
Conclusions • Quick process thanks to FFT (1 sec. for 220 × 320 pixels2 image, with 500 MHz Intel Celeron processor) • Image’s gradients method: more efficient. Works well with images that have faint and badly drawn areas. • With a good evaluation model, maximum error committed is 2.4% (1st method), and 1.3% (2nd method). • The best to reconstruct the unknown area of a registered image is to take this area identically in the reference image. • Can be used for a local purpose, on sections of images, choosing preferably side lengths that are power of 2 (FFT). &
17
%
$
'
References [Kuglin & Hines, 1975] C. D. Kuglin and D. C. Hines, “The Phase Correlation Image Alignment Method,” in Proc. IEEE 1975 Int. Conf. Cybernetics and Society, September 1975, pp. 163–165. [Foroosh, Zerubia & Berthod, 2002] H. Foroosh, J. Zerubia, and M. Berthod, “Extension of Phase Correlation to Subpixel Registration,” IEEE Trans. Image Processing, vol. 11, no. 3, pp. 188–200, 2002.
[Argyriou & Vlachos, 2003] V. Argyriou and T. Vlachos, “Sub-Pixel Motion Estimation Using Gradient Cross-Correlation,” in The 7th International Symposium on Signal Processing and its Applications (ISSPA), Paris, 1–4 July 2003.
&
18
%