Finding matching points between two 2d point sets, but different sizes

I am trying to find the way of identifying matching points between two sets (they are xy coordinates from two shifted images). Their sizes are different. Both sets have many points that are not shared. In other words, I look for algoriths to find same points between two shifted 2D point sets, which are not identical.

[weikang zhao]

How to define "matching points"? you should give a couple of legal points in your .mat file.

[photoon]

Hi Weikang,

row(xy1) row(xy2)

Point1 1 146

Point2 31 32

Point3 111 126

They are three of the matching points.

[weikang zhao]

What are the matching criteria? Is there a fixed distance between two points?

[KSSV]

How the given points are matching? Any logic? xy1 has two columns and xy2 has three columns.

[photoon]

I am sorry. Let me attach them again.

There are no logic. I just chose them visually. If you scatter plot them, you will be able to tell easily.

[photoon]

If we just use closest distance criteria, that wouldn't work because images from which points are collected are drifted from each other.

Answer 1

[weikang zhao]

if you dont need a very precise result,

clear
load('xy1.mat');
load('xy2.mat');
fixdistance=[-0.27,2.3];
newxy1=xy1+fixdistance;
dis=@(x,y) sum((x-y).^2);
result=struct([]);
structcount=1;
for i=1:size(newxy1,1)
    for j=1:size(xy2,1)
        if dis(xy2(j,:),newxy1(i,:))<0.1
            result(structcount).xy1num=i;
            result(structcount).xy2num=j;
            result(structcount).xy1=xy1(i,:);
            result(structcount).xy2=xy2(j,:);
            structcount=structcount+1;
        end
    end
end

the struct result contains the results you need. If you need a more general and more accurate method to deal with a large number of similar problems, you need to design an algorithm to estimate fixdistance.

have fun!

[weikang zhao]

by the way, optical flow may be help to estimate fixdistance

[photoon]

How to estimate fixdistance was my question. Thank you for your nice rephasement

Answer 2

[weikang zhao]

I provide a feasible solution. Assuming that the length of xy1&`xy2` are m and n, first generate a set of size m*n, including the distance between any pair of points, and then deploy a clustering algorithm or GMM fitting algorithm, the cluster center is the fixdistance.

[photoon]

Sorry, what is xy1&`xy2`?

[weikang zhao]

The two sets you provided

Answer 3

[KSSV]

Read about knnsearch.

[photoon]

I read knnsearch. Simply running knnsearch(xy1, xy2) gives a result. However, I don't thinkt this is what I want. Do I have a misunderstanding? Or could we apply this idea for measuring similiarity between two point sets?