The following product are based on IB3 Face Recognition Foundations, an extensive library of imaging and face recognition functions that along with training engines and SDK Generators allow to generate all the specific SDKs that are needed, such as:

  • IB3 Finder: Based on five pipe-lined algorithms specialized for fast localization of image areas that are likely to contain faces and the accurate localization of facial features such as eyes, mouth, eyebrows etc.
  • IB3 Recognizer: Researches the position of the face, normalizes the image, calculates the main features, formulates a statistical description of the face and then performs a statistical comparison of the similarity against a database.
  • IB3 Verifier: ISO 19794/5 Conformity Check evaluates facial images according to the ICAO ISO/IEC 19794-5 standard that defines the requirements for digital image geometry and scenery and returns Token Frontal Face Images and Full Frontal Face Images that are compliant with it.
  • IB3 Classifier: Performs a statistical recognition of people by gender, age, ethnicity etc. to aid in the classification of individuals of interest to the user.
  • IB3 Virtualiser: 3D Face Generation from one face image creates in real time a 3D facial model from a frontal facial image and generates a virtual face image with different poses that dramatically improves the performance of face identification in “non-collaborative” situations.
  • IB3 Custom SDK: Specialized SDK for specific customer oriented applications.


how to localise the face within the image

Localisation is based on two pipelined algorithms specialised for

  • Fast localisation of image areas that are likely to contain faces
  • Fast multi-scale and multi-location search

Normalisation of the acquired face

Accurate localisation of facial features (eyes, mouth, eyebrows etc.)

  • Subspace projection to optimise the parameter set
  • Robust against unfavourable illumination conditions
  • Robust against occlusions (dark sunglasses, scarves, etc.)

FRGC 2.0 test results

  • Good performance with medium-to-low facial image resolution (70-35 px eye-to-eye distance)
  • Good performance with uncontrolled facial image
  • Small template size 450-2250 FP array, 1.8-9 KB
  • Fast template generation 25 biometric templates/sec*
  • Fast similarity score generation 1.000.000 similarity score/sec*

*CPU Intel Xeon W3570 – 3.20Ghz

Feature extraction & storage

how to extract the features of a face

Proprietary tagged face image database built with a homogeneous criterion regarding gender, ethnicity and age

Tag information classes:

  • Position of Face Mask points (84 or 91 relevant points of measures
  • Photographic set: focus, illumination, background, etc.
  • Subject: gender, age, ethnic group, etc.
  • Face: pose, eye expression, gaze, mouth expression, etc.
  • Morphology: eye type, lip type, nose type, mouth type, etc.


how to verify the quality of the image as per ICAO/ISO 19794/5

Evaluates facial images accoring to ICAO/ ISO/IEC 19794-5, which defines the requirements for digital image geometry and scenery, and returns Token Frontal Face Images and Full Frontal Face Images that are compliant with it.

Supports the CBEFF Patron Formats A-C with ISO/IEC 19785-1 and ANSI INCITS 398-2005 interchange files, and its output is fully compliant with ISO/IEC 19794-5

Competitive advantages

  • Accurate and robust face finder
  • Accurate quality control check
  • Very fast processing: real-time ISO-compliant check & acquisition


how to get a virtual 3D out of a 2D image

3D model

  • creates in real time a 3D face model from a frontal face image
  • generates a set of virtual face images with different poses

The use of virtual face images may improve the performance of Face Identification task in “non-collaborative” settings

Enrolment process


how to recognise a collaborative person

  • Research of the position of the face: finds position and size of each one face in the image
  • Normalisation of the image: marks the points of morphological interest, as eyes, mouth, eyebrows, in order to properly scale and rotate the face placing those points in predefined positions
  • Calculation of main features: selects relevant features from the normalised image in order to maximise the robustness against environmental noise, non optimal pose and non neutral expression, beside varying lighting conditions
  • Statistical description of the face: a trained statistical engine processes the extracted feature vector template. The final feature vector is minimised to optimally describe the face
  • Similarity check on the database: measure of distances between feature vectors, or similarity value of the compared identities

Classifier – Audience intelligence

how to classify people of interest

Statistical recognition of people

  • Gender: Male or female
  • Age: chil, young, adult or senior
  • All ethnic origin: African, Asian, Caucasian

Attention time

Real watch count

Automatic generation of histogram