The law of relaxation is to allow the candidate match pair in T 
to dismiss oneself and to automatically match each other 
through iterative so as to make the “continuity" and 
"uniqueness" to obtain biggest satisfaction. The continuity 
refers to the massive other correct match pair usually existing in 
the neighborhood of correct match pair; Uniqueness refers to 
the identical feature point existing in only one matched pair. Or 
it can be expressed as the phenomenon that if candidate 
matching is right, there must be many candidate matching 
around it, while if candidate matching is wrong, there are less 
candidate matching around it. Matching support is defined as 
the degree that the neighbour candidate supports the candidate 
matching. It means that the strongest the matching support is, 
the more possible that the candidate matching is true.The 
detailed calculation is as below [2] ’ [3] : 
Suppose there are two feature points sets: P = {P],P 2 ,---P m } and 
Q-{Q\,Q2’ "Qn}i Define relative excursion between the two 
feature points sets for each paired points (Pj ,Q- ).Sjj(h,k) is 
relative distance between Pj,P^ and Qj,Q k when Pj and Q - 
partner (only shift). 
\d{P i ,P h )-d{Q j ,Q k i)\ 
S:j(h,k) = —- ———(/ = 1,2...) (1) 
lJ dist(P r P h -Q r Q kl ) 
Here: 
d(P h P h }= || Pj - P h || is the Euclid distance between /)• and P k . 
d(Q,,Qki) HI Qj -Qki II is the Euclid distance between Qj and Q kl . 
dist(Pi,P h \Qj,Qki) = [d{Pi,P h ) + d(Qj,Qki)]/2 is the average 
distance of the two pairing. Suppose | 8jj(h,k) = 0 | that means 
Qk relative to Qj and P k relative to Pj have the same meaning. 
So points (Ph ,Qk) should sustain (Pj, Qj ) as it the maximum. 
Along with | 8jj(h,k) | increase its support measure reduces. 
l+| ¿„(MO I 
When Pj partner Qj, Ph partner only with Q k .that is relative 
with Ph and the maximize support measure to ( Pj , Qj ) 
is Q k alone. The support measure from formula: 
Sjj(h,k)\) (3) 
max ( 
k*j 
-Sj(h,k)/e r 
0 
<K\Sjj(h,k)\) = ie 
Here: Qj is one of Pj matching candidate points, 
and\djj (h,k)\ < 8 r , (f>(\ 5jj(h,k) \)=e in other case 
<t>(\ Sjj(h,k) |) = 0. £ r is the threshold of relative distance change. 
When accounting use the experiential value. 
Because Ph does not have only one matching candidate 
point Qki , there will be more value of ^(| 5jj (h,k) |) . 
max (j)(\ 8jj (h, k) |) as the support measure of point Ph and its 
k*j 
matching points (Pj,Q¡). In the actual account, there is not only 
one point in adjacent field off*. If N(Pj) expresses the points 
set in adjacent field of Pj (without Pj), calculates the support 
measure that points of N(Pj) to points (Pj,Q¡) one by one. 
Finally the average value after accumulative is the total initial 
support measure: 
S°(Pi,Qj) = -Y max^fl 8jj(h,k) |) (4) 
m k* j 
h*i 
Where, m is the number of points in N{Pj). 
When calculating S°(Pj,Qj), every points (Ph,Qk) should be 
treated equally at first. Because there is no priori knowledge at 
beginning. After iteration for r times (r>0), the support measure 
that (Ph ,Qk) to (Pj, Qj) does not only relies on difference of 
position between P h and Q k , but also on their value of 
S r_l (Pj,Qj) which is the feedback of permission local support 
measure. The two factors can be combined together in different 
way. The least minimum is taken. 
S r (Pj,Qj) = — xY max mm[S r ~\Pj,Qj),<f>(\ Sjj{h,k) |)] 
m k*i J J 
This iteration continue until except the most possible point the 
support measure of rest points less than threshold which has 
already given to every P t . 
Stereo and sequence match simultaneously exist in the 3D 
feature correspondence movement analysis. The method and