The output is now the signed distance to the decision function. The classifier that yields the best results is to simply take the sign of the distance. Positive means legit, negative is labeled as an attack.
Positive distances may not purely be interpreted as a measurement of quaility whereas negative distances could be used as a tampering measurement, but, in an objective way since we are just measuring differences in the images.
The results might not agree with what a human would say. For that we would need to build an experiment where people score our tampered videos and compare the results with the scores of our function.
The output is now the signed distance to the decision function. The classifier that yields the best results is to simply take the sign of the distance. Positive means legit, negative is labeled as an attack.
Positive distances may not purely be interpreted as a measurement of quaility whereas negative distances could be used as a tampering measurement, but, in an objective way since we are just measuring differences in the images. The results might not agree with what a human would say. For that we would need to build an experiment where people score our tampered videos and compare the results with the scores of our function.