Can I use SSim to calculate the precision and recall for the detected model?

If I applied SSim to measrure the similarity between two images to detect crack , can I calculate the precision and recall by assuming threshold=0.5 so if ssim> or equal 0.5 is True postive ,and if

ssim<0.5 False positive

or precision and recall only for IoU measure?

Answer 1

[Shreeya]

A higher SSIM, which in your case is > 0.5 indicates that it is either a True Positive (Correct classification as similar images) or False Positive (Wrong classification as similar images).

If you have an access to the ground truth values, that is, 1 if the images are to be classified as similar and 0 if not, you can classify the prediction as a TP or FP and then calculate the precision using the following formula:

Recall can also be calculated similarly.