THE PHOTO VARIABLES / EVIDENCE INDEX
Facial asymmetry in photos
An uneven-looking photo or a changed symmetry score can reflect how the picture was taken or processed; it does not by itself show that your face changed.
For a guide that starts with capture conditions, read Why Does My Face Look Asymmetrical in Photos?.
This page brings our published experiments into one comparison table, followed by a short capture checklist. Each variable links to the original conditions, repeated results and limitations.
What we changed. What we measured.
These are observations from small, case-specific experiments, not a normal range for faces. Totals are displayed scores out of 100, not probabilities. No new experiment was run for this page; rows are not independent people or comparable accuracy tests. Rejections remain unscored.
| Variable | What we measured | What it means |
|---|---|---|
| Capture distance and head turn | Near/front: 98.6; far/front: 96.1, each repeated three times. Both turned captures were rejected in all repeats. | Different captures gave different outputs. Several capture details changed; distance alone was not isolated. |
| Image dimensions / resampling | 512 × 512: 98.0; 384 × 384: 97.8; 256 × 256: 98.1. Three matching repeats per size. | Resampling this portrait changed its displayed score; smaller did not consistently mean lower. |
| JPEG encoding | Baseline PNG: 98.0. JPEG quality settings 90 / 50 / 20: 97.9 / 97.7 / 97.5. Three repeats per input. | Encoding affected this input. These quality settings are not compression percentages. |
| Image-plane rotation | Baseline: 98.0; −10°: 97.9; +10°: 98.1. Three repeats per input. | This is a rotated image with resampling and canvas fill, not a simulation of turning a head. |
| Cropping | Baseline: 98.0. Removing 26 / 51 pixels from each edge: 97.6 / 97.9. Three repeats per input. | Changing the framing changed the output; this does not establish an ideal crop. |
| Mirror reversal | Original: 98.0; mirrored: 97.9. Mouth: 97.1 → 95.3. Three repeats per input. | The displayed detector score was not mirror-invariant in this case. The flip itself does not change geometric symmetry. |
| Page reload / repeatability | Three uploads after each of three page loads: all nine totals were 98.0, with matching feature scores. | Repeatability for this file and browser is not a measure of accuracy. |
| Placement on an added canvas | Left / center / right: all nine planned attempts returned “No face found.” Three later original-file controls each scored 98.0. | No score means no measured difference. The cause of the padded-input failures was not isolated. |
How to take a more comparable photo
- Keep the capture setup consistent. Use the same device, camera, camera height and distance between comparisons. Leave room to step back instead of placing the camera close to your face; our experiments do not establish a best distance.
- Face the camera directly. Keep your head upright, use a neutral expression and choose even light. A turned-head refusal is a request for a different input, not a low symmetry score.
- Keep the original file. Avoid comparing a cropped, resized or repeatedly JPEG-saved copy with an untouched image. Check feature scores as well as the total, and keep failed attempts in your record.
We measure photo-based geometry, not beauty, and this is not a medical assessment.
Try the face symmetry test ↗