|
 |
| |
| Facial ID System |
SecureMantra FACE ID face recognition system is composed of several components that are appropriate to any modern 2D face recognition system |
|
|
Face detector – capable of detecting any number of frontal or near frontal faces in grayscale images. |
| |
Every patch of the image is modeled by face primitives and classified whether it belongs to the class of possible face images (synthesized grayscale pictures below the image are most similar to face model). |
|
Face pose estimator and facial feature detector – capable of detecting facial features (eyes in our case) as well as estimating head rotation. This data is used by preprocessor to eliminate affine transformation, to normalize image contrast, etc.
|
|
|
|
|
|
|
|
Feature extractor. Several stages of photometric normalization are used for extracting of the unique facial features at different spatial frequencies. The most discriminating of them (learned from the huge internal training database of faces of different gender, age, race and appearance) are approximated by Gabor wavelets and saved in the template for fast comparison. Photometric normalization at one of the scales and positions of most discriminating facial features for the face at all scales |
|
Feature generalization. If several templates of the same face are available, information from all of them can be joined to cover the range of different appearances in one template for even more precise recognition. |
|
Feature matcher. The function that compares two templates searches for similar patterns of saved facial features. To calculate the similarity between two templates is almost as fast as comparing two strings. |
|
| |
| Screen Shot for Facial ID System |
| |
|
|
