Hi,

What is the most suitable way to compare (2D) vector fields of the same size?

Thanks in advance,

Rob

May be something like

d = distm(x,y); % squared Euclidean distances
e = sum(min(d’)) + sum(min(d’)); % make criterion symmetric

in which x and y are both nx2 matrices.

The above assumes that x and y have the same mean and scale.

Bob Duin

Hello Bob,

Thanks for your reply. The trouble I have with your measure is that it ignores the structure of the vector field. e.g. if x = [ 1 1; 1 1; 1 1 ] than y = [ 1 1; 1 2; 1 -1 ] and y = [ 1 1; 1+0.5*sqrt(2) 1+0.5*sqrt(2); 1+sqrt(2) 1+sqrt(2)] give the same similarity although the direction of the vectors is different.

By the internet I found a measure based on a weighted sum of the exp’s of the magnitude of the difference vector and the angular difference.("Similarity Measure for Vector Field Learning” Li,Shen) and a complicated one using a Clifford-algebra.("Clifford Convolution And Pattern Matching On Vector Fields” Ebling,Scheuermann ). I wondered if there are others and which one is better suited for matching/recognition purposes.

Firstly I want to use the measure to detect interest points in a field based upon a video sequence.

Regards and greetings,

Rob