what are the similarity measures between two signals

Dear all,

i am trying to implement Active Shape Model based segmentation. i have developed a model/template for the shape. Next, with the model , i have to search it in a given image.

the template/model consists of many points on the image, each associated with a profile at its location.

To compare the model profile with the profile generated in the test image, i am using euclidean distance as one measure. the other is the correlation points generated by normxcorr2() function.

both are not giving satisfactory answers.

Are there any other measures of similarity between a given template (a 1D signal) and another 1D signal. Please suggest.

Answer 1

[Aslak Grinsted]

There are really many different measures and they are good at different things (e.g. phase correlation).

The two measures you use are good if there is no scaling, shear, or rotation of the feature in the image relative to the template you have created. If that is the case then i would probably stick with normxcorr2 but pre-processing the images to emphasize the features you are looking for.

It would be helpful to see an example of the template and what you would like it to match.

Answer 2

[prashanth A]

the model contains the "profile" to be matched in signal.. the normxcorr2 generated the correlation factor. but corresponding point in the image is not the actual point which is matching it...

[prashanth A]

this is the other file