In some cases, the goal is to report people as stranger when they are not recognized.  This is common in case case where there is an area where only authorized personnel are allowed.

If the goal is to report on strangers, then its important to understand the difference in how SAFR characterizes a stranger vs. an unrecognizable face. 

• Unrecognizable is a face that did not meet a certain minimum criteria necessary for accurate facial recognition.   Attempting recognition below this criterial would generate an unacceptably high number of false matches.
• Stranger is a face for which recognition was attempted AND dod not meet a certain minimum criteria to be able to definitively claim that the face is not matching an enrolled face. 

By default, thresholds for Unrecognizable are lower than thresholds for Stranger classification.  While its possible to change this, it is generally not advised.

SAFR's thresholds are shown in the dialog below.  The labels have been highlighted to help show the groupings.

• Yellow highlighted fields are for recognition.  Below these values, face is considered unrecognizable and SAFR does not attempt to send the face to the database for matching
• Green highlighted fields are for stranges identification.  SAFR will not characterize a face as stranger when face quality is below these thresholds.

The settings above are default for surveillance use case where you want to be sure that a classification of stranger is accurate.  In cases where its important to identify strangers more aggressively, the learning / stranger settings can be reduced to near recognition settings as follows:

• Face size left at 120 (see below)
• Center Pose Quality set to 0.1
• Sharpness and Contrast set to 0.05

To avoid false matches, the face size should not be reduced.  A better strategy is to increased the detected face size and ensure the camera is zoomed into the region of interest so the number of pixels on the face is at least 600 PPM (Pixels per Meter) or 150 PPF (Pixels per Foot).  The higher the better.  Following table provides the width of view for different camera resolutions to achieve this pixel density.

If region of interest is wider than the width indicated above, use sufficient number of overlapping cameras to accomplish the same pixel density.