Overview
The face image assessment module provides metrics including head pose, smile detection, inter-eye distance, closed eyes, etc.
Face image quality issues to investigate:
Input
-
BQAT-CLI: Create a folder named
dataunder your working directory and put your face images in this folder. -
BQAT-API / BQAT-Stateless: Upload images via web API.
-
BQAT-GUI: Upload images via web page interface.
Supported file type: JPG, JPEG, JP2, BMP, PNG
Output
BQAT will produce quality metrics generated by processing engine selected in addition to input file metadata. The outputs will be stored as CSV via BQAT-CLI or JSON via BQAT-API.
Optional processing engines:
BQAT:
| Column | Description |
|---|---|
| file | Path to the input file |
| ipd | Inter-pupillary distance |
| confidence | Confidence level of face dectection (not quality score) |
| bbox_left | Left border of the face bounding box coordinates in pixels |
| bbox_upper | Upper border of the face bounding box coordinates in pixels |
| bbox_right | Right border of the face bounding box coordinates in pixels |
| bbox_bottom | Bottom border of the face bounding box coordinates in pixels |
| eye_closed_left | Left eye close or not |
| eye_closed_right | Right eye close or not |
| pupil_right_x | X coordinates of right pupil in pixels |
| pupil_right_y | Y coordinates of right pupil in pixels |
| pupil_left_x | X coordinates of left pupil in pixels |
| pupil_left_y | Y coordinates of left pupil in pixels |
| yaw_pose | Yaw in head pose direction |
| yaw_degree | Yaw in head pose degree |
| pitch_pose | Pitch in head pose direction |
| pitch_degree | Pitch in head pose degree |
| roll_pose | Roll in head pose direction |
| roll_degree | Roll in head pose degree |
| smile | Smile detected or not |
| glasses | Glasses detected or not |
| image_width | Width of the input image in pixels |
| image_height | Height of the input image in pixels |
OFIQ:
OFIQ engine is still in early stage of development, might be unstable, use with care.
| Column | Description |
|---|---|
| file | Filename of the input |
| quality | MagFace-based unified quality score measure |
| background_uniformity | Gradient-based background uniformity |
| illumination_uniformity | Illumination unformity by summing up the minima of the histograms of the left and the right side of the face |
| luminance_mean | Luminance mean measure computed from the luminance histogram |
| luminance_variance | Luminance variance measure computed from the luminance histogram |
| under_exposure_prevention | Under-exposure prevention by computing the proportion of low-intensity pixels in the luminance image to assess the abscence of under-exposure |
| over_exposure_prevention | Over-exposure prevention by computing the proportion of high-intensity pixels in the luminance image to assess the abscence of over-exposure |
| dynamic_range | Dynamic range computed from the luminance histogram |
| sharpness | Sharpness assessment based on a random forest classifier trained by the OFIQ development team |
| compression_artifacts | Assessment of the absence of compression artifact (both JPEG and JPEG2000) based on a CNN trained by the OFIQ development team |
| natural_colour | Assessment of the naturalness of the colour based on the conversion of the RGB presentation of the image to the CIELAB colour space |
| single_face_present | Assessment of the uniqueness of the most dominant face detected by comparing its size with the size of the second largest face detected |
| eyes_open | Eyes openness assessment based on computing eyes aspect ratio from eye landmarks |
| mouth_closed | Mouth closed assessment based on computing a ratio from mouth landmarks |
| eyes_visible | Eyes visibility assessment by measuring the coverage of the eye visibility zone with the result of face occlusion segmentation computed during pre-processing |
| mouth_occlusion_prevention | Assessment of the absence of mouth occlusion by measuring the coverage of the mouth region from mouth landmarks with the result of face occlusion segmentation computed on pre-processing |
| face_occlusion_prevention | Assessment of the absence of face occlusion by measuring the coverage of the landmarked region with the result of face occlusion segmentation computed during pre-processing |
| inter_eye_distance | Inter-eye distance assessment based on computing the Euclidean length of eyes’ centres and multiplication with the secant of the yaw angle computed during pre-processing |
| head_size | Size of the head based on computing the height of the face computed from facial landmarks with the height of the image |
| leftward_crop_of_the_face_image | Left of the face image crop |
| rightward_crop_of_the_face_image | Right of the face image crop |
| downward_crop_of_the_face_image | Bottom of the face image crop |
| upward_crop_of_the_face_image | Top of the face image crop |
| head_pose_yaw | Pose angle yaw frontal alignment based on the 3DDFAV2 |
| head_pose_pitch | Pose angle pitch frontal alignment based on the 3DDFAV2 |
| head_pose_roll | Pose angle roll frontal alignment based on the 3DDFAV2 |
| expression_neutrality | Expression neutrality estimation based on a fusion of HSEMotion with Efficient-Expression-Neutrality-Estimation |
| no_head_coverings | Assessment of the absence of head coverings by counting the pixels being labeled as head covers in the mask output by the face parsing computed during pre-processing |
BIQT:
| Column | Description |
|---|---|
| file | Path to the input file |
| background_deviation | Image background deviation |
| background_grayness | Image background grayness |
| blur | Overall image blurriness |
| blur_face | Face area blurriness |
| focus | Overall image focus |
| focus_face | Face area focus |
| openbr_IPD | Inter eye distance from openbr |
| openbr_confidence | confidence value from openbr |
| opencv_IPD | Inter eye distance from opencv |
| opencv_eye_count | Eye count from opencv |
| opencv_face_found | Face count |
| opencv_face_height | Height of face detected |
| opencv_face_width | Width of face detected |
| opencv_frontal_face_found | Number of front facing head found |
| opencv_landmarks_count | Landmarks of face detected |
| opencv_mouth_count | Number of mouth detected |
| opencv_nose_count | Number of nose detected |
| opencv_profile_face_found | Number of side profile of head |
| over_exposure | Overall image exposure value |
| over_exposure_face | Face area exposure value |
| quality | Overall quality score |
| sap_code | Sap code |
| skin_ratio_face | Skin to face area ratio |
| skin_ratio_full | Skin to fill image area ratio |
| image_area | Image area |
| image_channels | Number of image colour channels |
| image_width | Width of the input image in pixels |
| image_height | Height of the input image in pixels |
| image_ratio | Image aspect ratio |
| openbr_left_eye_x | Left eye x coordinate in pixels |
| openbr_left_eye_y | Left eye y coordinate in pixels |
| openbr_left_eye_x | Right eye x coordinate in pixels |
| openbr_left_eye_x | Right eye y coordinate in pixels |
| opencv_face_center_of_mass_x | Face center of mass x coordinate in pixels |
| opencv_face_center_of_mass_y | Face center of mass y coordinate in pixels |
| opencv_face_offset_x | Face offset x coordinate in pixels |
| opencv_face_offset_y | Face offset y coordinate in pixels |
| opencv_face_x | Face x coordinate in pixels |
| opencv_face_y | Face y coordinate in pixels |
| opencv_left_eye_x | Left eye x coordinate in pixels |
| opencv_left_eye_y | Left eye y coordinate in pixels |
| opencv_right_eye_x | Right eye x coordinate in pixels |
| opencv_right_eye_y | Right eye y coordinate in pixels |
| opencv_mouth_x | Mouth x coordinate in pixels |
| opencv_mouth_y | Mouth y coordinate in pixels |
| opencv_nose_x | Nose x coordinate in pixels |
| opencv_nose_y | Nose y coordinate in pixels |
Not all the columns are included in the table above for simplicity, for instance, there are normalized or scalar value of the same metrics.