How Selfbaus Reads a Face Photo: Gate, Measure, Report
Selfbaus is built around facial analysis and a practical routine you choose. This article explains the first layer, how a photo is checked and measured on your device. The routines and the evidence behind them are covered in the last articles of this section.
Most face-rating tools hand you a number and keep the method private. Selfbaus does the opposite. This article walks through what happens between the moment you choose a photo and the moment you see a report, and it marks the places where our method is a design choice rather than established science.
Step one: the gate
Before anything is measured, the photo has to pass a frame check. It needs to be a readable JPEG or PNG, contain exactly one verified face, be large enough, be sharp enough, not be badly under- or overexposed, and show a head pose close to frontal. If any check fails, you get a retake message and no score. The reason is simple: a measurement taken from an unusable input is not a low score, it is a meaningless one. Refusing to grade is more honest than grading.
Step two: deterministic measurements
When the photo passes, the engine measures it. Exposure, contrast, sharpness, noise, resolution and face framing come from pixel statistics inside the face box. Head pose (yaw, pitch, roll) and the 2D geometry values come from a face landmark model. Nothing here is a fitted "beauty" model. Each number has units and a defined method, and the same photo gives the same result.
Step three: a draft technical score
The 0-100 photo-quality score is a weighted blend of the technical measurements only. The weights are a draft heuristic of ours, not a result from a study. When a measurement is unavailable, it is left out and the remaining weights are rebalanced, so missing evidence never turns into a made-up number.
What the score is not
It is not a rating of your face. Willis and Todorov (2006) found that trait judgments made after a 100 ms exposure to a face correlated highly with judgments made without time limits. Jenkins and colleagues (2011) showed that photographs are blind to within-person variability, which is often very large compared with differences between people. A single photo score therefore says something about that photo. Use it to compare your own shots taken under similar conditions.
Where each part is explained
The rest of the Science section covers the face mesh, head pose, camera distance, photo quality checks, research on perceived health cues, what routine guidance says, iris-based normalisation with the three ratios, symmetry, and proportions.
References
- Willis, J., & Todorov, A. (2006). First impressions: making up your mind after a 100-ms exposure to a face. Psychological Science. https://pubmed.ncbi.nlm.nih.gov/16866745
- Jenkins, R., White, D., Van Montfort, X., & Burton, A. M. (2011). Variability in photos of the same face. Cognition. https://doi.org/10.1016/j.cognition.2011.08.001