Hi, In connection with pattern identification in images, cross-correlation between the image and a template matrix is often used. Is there any reason why the means are substracted from the elements before the multiplying them? If the only purpose is to detect similarity, the scaling seems unessecery. Thanks in advance
Normalization in cross-correlation (image processing)
Started by ●April 21, 2008
Reply by ●April 21, 20082008-04-21
Simon Johan wrote:> Hi, > > In connection with pattern identification in images, cross-correlation > between the image and a template matrix is often used. Is there any reason > why the means are substracted from the elements before the multiplying them? > If the only purpose is to detect similarity, the scaling seems unessecery.Subtracting the means is not the same as scaling. Some operations (integration, for instance) are upset by a non-zero mean. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●April 21, 20082008-04-21
>Hi, > >In connection with pattern identification in images, cross-correlation >between the image and a template matrix is often used. Is there anyreason>why the means are substracted from the elements before the multiplyingthem?>If the only purpose is to detect similarity, the scaling seemsunessecery.> >Thanks in advance > > >Actually, these calculations are applied to general cases. Image that you two matrices which stand for two individual pictures. The number elements of first matrix are range from 0 to 1, while the numbers elements of the second matrix are range from 30 to 251. Maybe the pictures are captured from exactly the same sense and content are completely the same, just with different illumination etc. What you are supposed to do with these two matrices, if you want to determine the similarity? So normalization is always an important step for both pre-processing and finalization.
Reply by ●April 21, 20082008-04-21
There are several tricks for making this algorithm work better. Removing the mean is one of these. Here is a recent blog that shows the importance of these tricks. Regards, Steve http://www.dsprelated.com/showarticle/52.php