Analyze Faces

Face analysis estimates the demographics and/or state (expression) of a face. Face demographics include classifying the face’s apparent age, ethnicity, and gender, while face state estimates the expression of the face. Media Server can return any of the following results for each face.

  Age Ethnicity Gender Expression
Possible Results
  • Baby
  • Child
  • Young Adult
  • Adult
  • Elderly
  • Arab
  • Black
  • East Asian
  • Latino
  • South Asian
  • Southeast Asian
  • White
  • Female
  • Male
  • Unclassified
  • Happy
  • Neutral
  • Unknown

Each face that is analyzed is detected using the face detection module.

NOTE: Estimation of ethnicity is based on face shape only, and not on skin color.

Configure Face Analysis for Best Results

Face detection is the first step in face analysis, therefore all of the parameters used to tune face detection also apply to face analysis.

Factors Affecting Face Analysis Performance

The accuracy of face analysis depends largely on the images that it processes. Estimating face demographics and expression is a complex operation and therefore requires a good quality input image. The factors that affect the performance of face analysis are the same as the factors affecting face detection, only more restrictive.

  • Face direction. It is very important that the face is looking directly at the camera. Both eyes must be visible, and as much of the cheek and jaw bones on both sides of the face should be as visible as possible.
  • Face size. The face should be large and form a large proportion of the image.
  • Facial expression. Although Media Server can analyze faces with different expressions, facial expressions that cause a larger scale change in face shape (for example, winking or gaping) inhibit demographic classification.
  • Lighting. Lighting should be consistent across the face, with no areas of shadow.