Face Quality Assessment

Nowadays there are several state-of-the-art technologies which allow to detect objects inside an environment. Nevertheless, mainly the uncontrolled acquisition of faces involves problems such as: focusing the target, keeping right illumination conditions and identifying best facial poses. All these aspects affect performances in terms of picking images which are suitable for a human operator and useful for automatic identity recognition systems.
This application implements an innovative set of methods and algorithms, which are able to perform the assessment of faces captured from a certain subject, based on their quality.
In this context, the concept of quality is quantified by measuring the two main critical features, that's the sharpness of the image and the degree of frontality of the face.
In order to estimate sharpness level, this tool makes use of a new algorithm which is able to automatically detect the best set of metrics among 12 totally applied on the single image. Each metric implements a different state-of-the-art approach in sharpness evaluation.
The optimal set of metrics that will be used to evaluate the whole sharpness score, is detected each time a new different gallery has been captured for a certain subject.
In order to perform pose estimation instead, this tool makes use of a new geometric approach based on definition of a semi-rigid model of frontal face, which uses golden section in order to approximate proportions among various facial points. The amount of frontality is quantified by measuring the Euler angles (Roll, Yaw, Pitch), that in this case represent the orientation of the head in the three-dimensional space.
Once degree of frontality and amount of sharpness for each image in the face gallery are computed, so the application can perform another important operation, that's the assessment of the gallery in terms of face quality.
The whole quality score is obtained by suitably combining three different parameters: frontality, sharpness, and size of the face.

Category:

Science & Technology

Tags:

• face
• assessment
• quality
• sharpness
• 3D
• head
• pose
• estimation
• computer
• vision
• opencv
• Amilab
• amibient
• intelligence
• videosurveilance
• Intelligenza
• Ambiente
• videosorveglianza
• stima
• posa
• volto
• qualità
• ordinamento
• nitidezza