Hi,

If arcos is arccos (cosine inverse), then this should represent the Pearson Correlation Coefficient between descriptor vectors A and B. If A and B are two vectors, and mA and mB are two values (also vectors) of A and B, respectively, then mA.mB represents the dot product of vectors mA and mA, and the formula represents the geometric interpretation of Correlation Coefficient.

Source: http://en.wikipedia.org/wiki/Pearson_product-moment_correlation_coefficient

Dear Hadjer,

I agree with @Dibyendu regarding the geometric interpretation.

Please note that mA and mB are not "values". They are vectors representing descriptor A and B, respectively. These descriptors do not have to be vectors.

The geometric definition of the dot product mA.mB is

mA.mB = ||mA||.||mB||.cos(T) where T is the angle between mA and mB.

Then, by extracting T from the above, you get your formula which gives you therefore the angular separation T between mA and mB.

In your case, if T is close to pi/2 or -pi/2 then the interpretation is that the descriptors A and B do not share attributes.

Also note that, unless you need specifically the angle T, it is perhaps better and faster if you do not compute the inverse cosine but just the argument inside your formula. You will get cos(T) instead of T and you will obtain then the standard cosine distance with values between -1 and 1.