0
$\begingroup$

I am following the tutorial to calculate the HOG feature matrix. However, it does not tell how to find the object using the HOG matrix.

For example, I have an image of several pipes stacked together. I have a template image of one pipe. I want to find out the total number of pipes in the original image. I don't have training images to train the SVM. I have just one template image. Using the HOG of the template image I want to find the total number of pipes in the original image. Can it be done?

I want to know if it be done without using SVM and true positive images without having to train the SVM model.

$\endgroup$

1 Answer 1

2
$\begingroup$

You need to compute the distance (cosine distance for instance) between the descriptors in both images in order to match them.

So for instance, compute the HOG descriptor of a pipe and try to match it with every cells in the other image. You need a threshold too to decide if two descriptors are matched or not.

However HOG descriptors are not rotation invariant so you probably need descriptors for different orientations of one pipe.

$\endgroup$
1
  • $\begingroup$ how to do this step --> match it with every cells in the other image $\endgroup$ Sep 5, 2023 at 4:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.