1-1-4 
it takes 2 log —bits (Forstner, 1989; Vosselman, 1992). 
£ 
Thus 
mp„L)=S'(0 =2/v «> gj 
i„ u , =1 t 
(9) 
On the other hand, conditional information has been 
defined in Ingels, 1971; Blahut, 1987 by conditional 
probability as follows. 
I(a i \b j ) = -\og?vob(a i I b¡) (10) 
Thus, 
/.(£ !/’,)= I -l°g< 
<m=l)e/i 
27T<7 
(jr* ~/l(xf m )) 2 +( Vj )) 2 
2a 2 
) 
(11) 
Since a matched point should use fewer bits in description 
than an unmatched one, it has 
2 (x)-h(xf)) 2 +(y'j-h(yì)) 2 
-log 
27TC7 2 
<21og- 
£ 
Let 
= 2(7 Jin 
R 
yfbra 
(12) 
Image points r) = (x), y))‘ and r 2 = (xf, yf) T are 
matched ones, if and only if it has 
(13) 
B. Conditional information of line segments 
image points, angle between line normal and difference 
between orthogonal distance. Thus 
l h {l-\l)) = l h iP-\Pj) ( 16 ) 
+ I h (angle(n), nf )) + I h (dis(d), df )). 
Suppose A0* and Ad k s be the estimation error of n k s 
and d k s in l k (s=i,j; k=l,2), following normal 
distribution, then it has 
angle 2 (!) Mlf)) 
I h (angle(n],nf)) = - log( 
yfïna. 
) 
(17) 
I h (dis(d),d?)) = - log( 
dis 2 {!) Ml?)) 
2a d , ' 
) 
dij 
(18) 
Where 
=V (<T « )2 +( <J . 2 ,) 2 ' 
=^jj) 2 +(^,-) 2 • 
Since a matched line segment should use fewer bits in 
description than an unmatched one. Thus, line segments 
l'. and if are matched, if and only if it satisfies 
/»(/, 2 l/;)<2W i 2 log-=i i 2 . (19) 
£ 
Given a mapping h from d 2 to D,, L 2 is divided into 
two groups. 
1 ) L l„ = Vl 1 c = ) is the group of line 
segments having no match in L,. 
2) Lf n '= {if I i in = } is the group in which each 
line segments if have a match l\ in L,. 
Thus, 
K (/? 2 1 R x ) = I h (L 2 1 L x ) = /„ (L 2 0Ut ) + l h (Li I L,) 
(14) 
Line segments without match can be represented using 
their point primitives. Thus, 
$oui D 
/»aL)=2£w,io g — (is) 
r =1 £ 
l oul 1 
Where, w,. is the number of image points in 1.. 
Conditional information between two matched line 
segments is analyzed by three components, match in 
2.1.4 Searching for best matching 
A. Coarse matching 
Searching for coarse matching is conducted in an 
exhaustive mode. Searching space concerns with the 
angle between non-parallel line segments and distance 
between parallel line segments. Line segment pairs with 
the same intersection angle or distance (if parallel), are 
tried to match the images. The best match is selected to 
be the one yielding minimum distance measure. It set the 
initial value to the transformation between two images. Let 
it be T°. 
B. Strength definition in fine adjustment 
In fine adjustment, strength between each matched pair of 
line segments are analyzed to lead the adjusting 
procedure. Let T k be a transformation from D, to D,.