A threshold is a balance between false positives and false negatives (not recognising someone whom you are supposed to recognise). Searches (and verifications) return a Similarity Score. The threshold is a Similarity Score (selected by you) above which you will accept a search/or verification result to be the same person.
The threshold setting recommended by Seventh Sense is 0.685, but it is more use-case dependant.
This means:
1. If you select a threshold of 0.7, False Positives will decrease, False Negatives will increase, and overall accuracy will decrease.
2. If you select a threshold of 0.65, False Positives will increase, False Negatives will decrease, and overall accuracy will decrease.
In scenario 1, it will be more challenging to identify someone (it might take longer as the majority of sampled faces won't match), but when a person is recognised, you can be more confident that it is the correct person.
In scenario 2, it will become easier to recognise someone (it will be quicker as fewer sampled images will fail to match), but you may encounter false positives (similar faces of two different people may match).
Each use case determines the threshold setting. For example, in access control, you might set a higher threshold to ensure that only the right people are allowed entry. However, you might keep a lesser threshold or the optimum threshold point in a counter-terrorist operation as you may not want to miss a .