Playground biometrics demo Playground BioID home page Company Website

OFIQ Quality Measures

You can explicitly perform quality checks by calling into the face image quality measurement implementation of the ISO/IEC 29794-5:2025 standard. This implementation is based on the Open Source Face Image Quality (OFIQ) library. For more information, see the official OFIQ GitHub Project.

The table below lists the 28 OFIQ quality measures currently performed by our service in the order as they are defined in ISO/IEC 29794-5:2025:

Quality Measure Description
GeneralGeneral prediction of the quality of an input image for face recognition.
UnifiedQualityScore A unified quality score (based on the MagFace50_FP16 CNN model) which aims to predict the utility of the facial image for recognition purposes. It is used to ensure that only facial images of sufficient quality are fed into a face recognition system.
Capture-related The following quality elements are related to the imaging system and acquisition environment. If a quality is evaluated as low, operators should check their system accordingly.
BackgroundUniformity Measures the mean squared lengths of the luminance gradients on the background of the face image. It utilizes a face parsing segmentation map and is mainly relevant for reference images for ID documents. (See also: ISO/IEC 39794-5:2019 Annex D.1.4.2.5)
IlluminationUniformity Measures the uniformity of lighting across the face by evaluating the intersection of luminance histograms in specific left and right measurement zones on the cheeks. It assesses conformance to requirements to prevent strong shadows that negatively affect recognition utility. (See also: ISO/IEC 39794-5 Annex D.1.4.2.6)
LuminanceMean Computes the mean luminance (brightness) within the landmarked facial region. It addresses exposure related defects known to negatively affect the reliability of biometric decisions.
LuminanceVariance Computes and outputs the luminance variance within the landmarked region. It measures the spread of brightness values to identify potential contrast issues.
UnderExposurePrevention Determines the proportion of pixels in the non-occluded face region with luminance in the range [0; 25]. It identifies images that are too dark, which can leave textures less visible for biometric recognition.
OverExposurePrevention Determines the proportion of pixels in the non-occluded face region with luminance in the range [247; 255]. Over exposure represents a capture-related defect that washes out texture details and negatively affects image utility.
DynamicRange Computes the entropy of the normalized luminance histogram in the landmarked region. It estimates the range of brightness levels used, which directly influences recognition performance. (See also: ISO/IEC 39794-5 Annex D.1.4.2.8)
Sharpness Assesses image sharpness using a Random Forest Classifier trained on Sobel and Laplacian filter outputs. High sharpness is required to ensure that texture details are visible to face recognition algorithms. (See also: ISO/IEC 39794-5 Annex D.1.4.2.4)
CompressionArtifacts Predicts the Peak Signal to Noise Ratio (PSNR) of the input image relative to an uncompressed source using a CNN model. It identifies blocking artifacts resulting from heavy compression (e.g., JPEG, JPEG2000) that may impair recognition performance.
NaturalColour Evaluates typical skin colour values in the CIELAB colour space within specific measurement zones on the cheeks. It detects unnatural colour casts or saturation which can negatively affect the utility of facial images. (See also: ISO/IEC 39794-5 Annex D.1.4.2.9)
Subject-related Related to subject behaviour. If a quality is evaluated as low, feedback should be provided to the biometric capture subject.
SingleFacePresent Ensures that exactly one face is detected in the image as required for biometric processing.
EyesOpen Measures the ratio of the maximal distance between upper and lower lid divided by the distance between eyes’ midpoint and chin. It assesses eye openness, as closed eyes can massively increase face recognition error rates. (See also: ISO/IEC 39794-5 Annex D.1.4.3.3)
MouthClosed Measures the ratio between the mouth's aperture and the distance between eyes’ midpoint and chin. It ensures conformance to requirements for passport style images where a closed mouth is typically required. (See also: ISO/IEC 39794-5 Annex D.1.4.3.2)
EyesVisible Computes the proportion of occlusion of the Eyes Visibility region using face occlusion segmentation. It checks if the eye region is free from occlusions, which are defects known to negatively affect recognition decisions. (See also: ISO/IEC 39794-5 Annex D.1.4.3.3 and ISO/IEC 39794-5 Annex D.1.4.3.4)
MouthOcclusionPrevention Computes the proportion of occlusion of the mouth region using landmark based segmentation. This specific component is used to detect defects like face masks or hair in the mouth area that hinder recognition. (See also: ISO/IEC 39794-5 Annex D.1.4.3.5)
FaceOcclusionPrevention Measures the proportion of the landmarked face region marked as occluded in a segmentation map, assessing the overall degree of occlusion by objects or hair to provide transparent, actionable feedback. (See also: ISO/IEC 39794-5 Annex D.1.4.3.5)
InterEyeDistance Estimates the inter-eye distance accounting for perspective projections due to the head pose. This measure evaluates the image resolution, which is essential for ensuring high sample quality. (See also: ISO/IEC 39794-5 Annex D.1.4.2)
HeadSize Calculates the distance T between the eye midpoint and the chin relative to the image height. It helps ensure the face size meets standard geometric requirements for ID and reference images. (See also: ISO/IEC 39794-5 Annex D.1.4.4)
LeftwardCropOfTheFaceImage Assesses the leftward crop measure based on the position of the right eye center relative to the inter-eye distance, ensuring sufficient margin exists on the left side of the face. (See also: ISO/IEC 39794-5 Annex D.1.4.4)
RightwardCropOfTheFaceImage Assesses the rightward crop measure based on the position of the left eye center relative to the inter-eye distance and image width, ensuring sufficient margin exists on the right side. (See also: ISO/IEC 39794-5 Annex D.1.4.4)
MarginAboveOfTheFaceImage Assesses the upward crop measure based on the distance from the eyes' midpoint to the upper image border, relative to the vertical face size (distance between eyes and chin). (See also: ISO/IEC 39794-5 Annex D.1.4.4)
MarginBelowOfTheFaceImage Assesses the downward crop measure based on the distance from the chin to the lower image border, relative to the vertical face size. (See also: ISO/IEC 39794-5 Annex D.1.4.4)
HeadPoseYaw Estimates the head rotation around the vertical axis (yaw angle) using a 3DDFA-V2 CNN model. See also ISO/IEC 39794-5 Annex D.1.4.3.1, which requires that the yaw angle of the head is less than ±5° from frontal.
HeadPosePitch Estimates the head rotation around the transverse axis (pitch angle) using a 3DDFA-V2 CNN model. See also ISO/IEC 39794-5 Annex D.1.4.3.1, which requires the pitch of the head to be less than ±5° from frontal.
HeadPoseRoll Estimates the head rotation around the sagittal axis (roll angle) using a 3DDFA-V2 CNN model. See also ISO/IEC 39794-5 Annex D.1.4.3.1, which requires the roll of the head to be less than ±8° from frontal.
ExpressionNeutrality Determines whether the facial expression is neutral using a dedicated CNN-based approach (HSEmotionWithTwoCNNs). Non-neutral expressions (e.g., smiling, squinting) distort facial topology and reduce the accuracy of face recognition. (See also: ISO/IEC 39794-5 Annex D.1.4.3.2)
NoHeadCoverings Determines the proportion of pixels in the face parsing segmentation map labeled as "hat" or "clothing" above the eyes. This detects items like hats or veils that might obscure the face, which is relevant for conformance to ID document standards. (See also: ISO/IEC 39794-5 Annex D.1.4.3.5)