iOS Recognition Preferences
Use Recognition Preferences to configure a variety of settings that determine whether or not SAFR detects, tracks, and recognizes faces and identities.
- Detect identity: Enables identity detection.
- Detect occlusion: Obstructing the full view of the face by using, for example, a mask, glasses, or using a hand to block a part of the face.
- Detect mask: Enables the detection of masks.
- Mask Model: Specifies the model to be used for mask detection.
- precise: This model produces the least number of false positives (i.e. detecting that a person is wearing a mask but there is no mask), but it suffers from the lowest true positive rate. (i.e. detecting masks that are actually there)
- sensitive: This model produces the highest true positive rate, but it suffers from the highest number of false positives.
- normal: This model produces a moderate amount of both false positives and true positives.
- Mask Threshold: Specifies the threshold at and above which mask detection will conclude that mask=true.
- Consecutive confirmations required: Adds additional required consecutive confirmations for SAFR to recognize faces covered by masks. Increasing this setting creates more reliable but slower facial detection.
- Detect gender: Enables gender detection.
- Detect age: Enables age detection.
- Detect sentiment: Enables sentiment detection.
- Detect smile action: Enables smile authorization.
- Pre-smile delay (seconds): The amount of time, in seconds, that there should be no smile.
- Smile duration (seconds): The amount of time, in seconds, that the smile should last.
- Identity Threshold Boost: The smile threshold to boost temporarily during the smile action.
- Transition thresholds: Specifies what range of sentiment values are classified as happy, neutral, and unhappy.
- Detect pose liveness action Enables pose liveness detection
- Center pose quality: Minimum center pose quality a subject must have before a pose liveness check is attempted.
- Profile pose quality: The maximum center pose quality to use when detecting the final profile pose.
- Max profile confidence at start: Advanced setting. Adjust only at the direction of SAFR technical support.
- Min profile confidence at end: Advanced setting. Adjust only at the direction of SAFR technical support.
- Min profile pose yaw: Advanced setting. Adjust only at the direction of SAFR technical support.
- Center pose consecutive confirmations: Number of consecutive center pose confirmations required before pose liveness is confirmed. Increasing this value causes pose liveness checks to take longer to complete and be more difficult to complete successfully, but you'll be more certain that people passing the checks aren't spoofed faces.
- Profile pose consecutive confirmations: Advanced setting. Adjust only at the direction of SAFR technical support.
- Min profile similarity: Advanced setting. Adjust only at the direction of SAFR technical support.
- Min detections per second: Advanced setting. Adjust only at the direction of SAFR technical support.
- Min transition poses: Advanced setting. Adjust only at the direction of SAFR technical support.
- Max CPQ jump in continuous tracking: Advanced setting. Adjust only at the direction of SAFR technical support.
- Max CPQ jump after tracking loss: Advanced setting. Adjust only at the direction of SAFR technical support.
- Max profile pose roll: Advanced setting. Adjust only at the direction of SAFR technical support.
- Detect mask action: Enables the detection of mask event types. Mask event detection attempts can return 3 potential results: mask=false, mask=indeterminate, or mask=true. After the specified number of consecutive mask event detection results, (configured below) the mask event state is set to the appropriate value. The mask event state can only progress from false towards true; the state never regresses back towards false. For example, once the mask event state for a viewed person becomes set to mask=true, then that person's mask event state won't ever regress to mask=indeterminate or mask=false.
Events are generated when the mask event state is set to either mask=false or mask=true.
Note that enabling this setting automatically enables the Detect Occlusion setting above, since detecting mask actions inherently involves detecting occlusion at the same time.
- Min Face Size: The smallest face size, in pixels, upon which the Mobile Client will attempt to detect a mask.
- Min Consecutive Mask Detections: Specifies the minimum number of consecutive mask=true mask detection results that must occur before the Mobile Client will generate a mask=true event.
- Min Consecutive No Mask Detections: Specifies the minimum number of consecutive mask=false mask detection results that must occur before the Mobile Client will generate a mask=false event.
- Min Consecutive Occluded No Mask Detections: Specifies the minimum number of consecutive mask=indeterminate mask detection results that must occur before the Mobile Client will set the mask event state to mask=indeterminate.
- Face Edge Threshold: How far a face must be from the edge of the screen before a mask event detection is attempted. For example, if a face is 100 pixels, and Face Edge Threshold is set to 3%, then the face must be 3 pixels from the edge of the screen before SAFR will attempt a mask event detection.
- Face quality settings:
- Minimum required center pose quality
- For recognition: Defines the minimum required quality for a face posed directly in front of the camera (center posed) to attempt recognition. Center pose quality (CPQ) ranges from one to zero. A score of 1 is given to a face looking straight into the camera. Any deviation from this position diminishes center pose quality. Center pose quality of a face in full profile position is given a score of zero. Recognition from any pose is possible, but accuracy is reduced for faces that are in extreme profile positions.
- For merging: Defines the minimum required face center pose quality to attempt merging with existing reference images for a recognized identity.
- For learning/strangers: Defines the minimum required face center pose quality to enable SAFR to store a reference image for a new identity.
- Minimum required face sharpness quality
- For recognition: Indicates minimum required face sharpness quality to attempt recognition.
- For merging: Indicates minimum required face sharpness quality to attempt recognition.
- For learning/strangers: Indicates minimum required face sharpness quality to store as a reference for a new identity.
- Minimum required face contrast quality
- For recognition: Indicates the minimum amount of contrast quality (lower or higher contrast) for SAFR to attempt a recognition.
- For merging: Defines the minimum required face contrast quality to attempt merging a captured face with its existing references in the SAFR system.
- For learning/strangers: Indicates the minimum required face contrast quality to store as a reference for a new identity.
- Minimum recognition face size: Defines the minimum required face size in pixels to attempt recognition. It includes a 25% margin around the face.
- Minimum learning face size: Defines the minimum required face size in pixels to enable SAFR to store a reference image for a new identity. It includes a 25% margin around the face.
- Minimum recognitions to lock on to identity: Minimum number of consecutive recognition attempts that must produce the same identity before SAFR locks onto the identity.
- Re-recognition delay: The number of seconds that must pass before SAFR attempts to reconfirm a tracked person.
- Initial recognition attempts: Number of initial recognition attempts to make on an unrecognized face as quickly as possible.
- Failed recognition back-off interval: After making the initial recognition attempts as quickly as possible, back up the amount specified by this setting for each subsequent recognition to slow down. This goes on until the retry interval is reached.
- Retry failed recognitions after every: The interval in which to run recognition requests if the face has not been recognized.
- Lingering timeout: Specifies the number of additional frames that SAFR keeps a tracked face in active memory after SAFR has failed to detect it in the most recent frame. Increasing this value makes the tracker resilient against intermittent loss of face, but consumes system processing power.
- Update identity every: Updates the identity when the currently saved identity is older than the updated identity.
- Update identity with better image: Updates the identity when the currently saved identity is of lower quality (in all aspects) than the new image.
- Maximum allowed occlusion:
- For learning/strangers: Indicates the maximum occlusion value allowed for a face to be registered to the Person Directory. When this setting is set to 1, no occlusion filtering is applied, and the default configuration is ignored.
- Learn occluded faces: Enables the learning of occluded faces.
- Identity Recognition Threshold: Determines the strictness of the face recognition when declaring identity matches between a face and stored identity image. You can independently set the Identity Recognition Thresholds for the following:
- Camera: Sets the threshold for images. (e.g. photos)
- Similar: Sets the threshold for Similar comparisons when running the Similar operator mode.
- Masked face threshold offset: Sets the threshold when attempting to recognize faces that are wearing masks.
- Proximity Threshold Allowance: A boost value that is added to the Identity Recognition Threshold.
For detailed information about Identity Recognition Threshold and Proximity Threshold Allowance, see Identity Recognition Thresholds.
- Max Concurrent Recognitions: Specifies the maximum number of concurrent recognitions that the Mobile Client can attempt. We strongly recommend that you do not change this setting; doing so could greatly decrease your device's performance.