Training Image Guidelines

Media Server creates a model of each face to recognize in a database.

You can provide multiple training images for each face model; however, in most cases Micro Focus recommends providing only one good quality image of a face that best matches the expected appearance of the face in recognition tasks. If you expect a face to show considerable variation, for example a person who occasionally wears spectacles, it can be beneficial to use multiple training images. Keep the number of additional training images to a minimum, focusing on providing as few good quality representations of the face as needed to cover all anticipated recognition scenarios.

It is important to use good quality training images to populate the database. A training image for a face:

  • must contain only one face.
  • must be rotated correctly (the face cannot be upside down).
  • contains a face that is wider than approximately 150 pixels; the face should constitute a large proportion of the image.
  • contains a front view of the face, so that both eyes are visible.
  • is not blurry.
  • has even illumination, because dark shadows or bright highlights on the face reduce recognition accuracy.
  • is free from compression artifacts (which can affect some image formats, such as JPEG images).

  • shows a face that is not partially obscured (for example a person standing behind a microphone).