|
Most of the automated machine solutions developed presently are based on intensity images or 2D. One clear difficulty in the approaches using still intensity images is that features are extracted using image contrast. First, many features of the face are difficult to detect or measure because of variability of light conditions or low dynamic range in the input image. Even the highest contrast features of the face, such as the eyes, are a challenge to identify and describe correctly. While low contrast features, such as the shape of jaw boundaries, cheeks, and forehead are currently impossible to describe precisely from general intensity images. This makes 2D face systems extremely sensitive to change in ambient lighting conditions between enrollment images and subsequently captured images. Another example of variance under illumination would be images of a person with or without makeup.
Secondly, 2D facial recognition systems work only by comparing ratios between key features. Such systems are not able to extract accurate or ground-based measurements of the facial structure. 2D images are therefore not flexible with respect to orientation or scale change. Performing a more reliable recognition requires the storage of multiple views or “templates” over varying conditions.
Therefore, even without changing any basic biometric properties of a person, 2D systems may perform quite poorly in real life scenarios.
TakSec VisionAccess Enrollment Stations
 Taksecs’ Facial Enrollment Station performs user enrollment and builds 3D template databases, as well as creating 2D color jpeg files and video feed, in accordance with ICAO standards. Designed with Taksec’s advanced patented optical technology, structured light, and algorithms and easily integrating into a range of existing hardware or software security systems, the enrollment station comprises a 3D camera and 2D color digital video camera. The outputs of the enrollment camera from a single enrollment shot are: a) a sequence of 2D color JPEG images (video feed) according to ICAO specifications; b) 3D image or reconstructed mesh; c) 3D template; d) 3D hologram; e) 2D black & white JPEG image for 2D matching engine. The output may either be stored in a database or written to a card or chip. This allows multiple outputs to be generated from a single enrollment shot in a fast and cost-effective manner. Taksec 3D data is stored in the standard Common Biometric Exchange File Format (CBEFF). The data described by CBEFF includes Security (Digital Signatures and Data Encryption, Processing information and the Biometric data. The 3D biometric data resides within the CBEFF structure is stored in a propriety encrypted format. Taksec’s efforts to protect the 3D data in conjunction with a strong network security policy by the customer provides a secure environment for the solution.
Robustness to Lighting & Angles
 Current 2D face recognition systems are based on standard photo or video pictures (intensity images) of a subject’s face – these are not ground-based measurements and are highly sensitive to changes in ambient lighting or view angle. In addition, they are sensitive to changes in scale, facial accessories (make-up, glasses, and beards) and aging of the user.
By contrast, VisionAccess technique uses direct geometric measurements available from 3D images (Range images) with the potential to also overlay a 2D texture map (3D + 2D). This 3D geometry is invariant to ambient light conditions and may even be used in the darkness, thanks to the near infrared light projector. In addition, 3D face recognition technology is more robust to different view angles between the enrolment & captured shots, with robust recognition up to 30° from the frontal position. 3D face has the future potential to recognise a user even at 90°, for surveillance or law enforcement applications. Therefore, this 3D approach has the potential to work with higher accuracy in real world environments.
Anthropometric Information
3D face biometrics is based on anthropometric data – precise measurements of the cranial structure & rigid tissues. The 3D biometric template (feature vector) is extracted from information about the cranial curvature in those areas where the rigid tissues are most evident (for example, eye sockets, superciliary arches, chin zone, bridge of nose). These areas are the most unique and are unchanging over time, and therefore robust to aging or weight changes in the subject.
Semantic Analysis
The key facial features are not only measurable, but also set in a known structure (the semantic analysis of the face). This is able to be visualized in 3D, and this permits a number of key functionalities in the software.
Firstly, the semantic analysis of the face permits the use of “smart” algorithms for facial camouflage. For example, in the event of the user growing a beard, the expected shape of the chin will be deformed by the hair and the 3D reconstruction will show a rough area of the face. This permits the algorithms to recognize that the user has grown a beard, and automatically shifts the emphasis of recognition (matching) to the alternate reliable areas of the face. The same system is used for glasses/spectacles, where the algorithms register a no concave area around the eyes. Depending on the level of security required, the software may be programmed to request the user to remove the glasses, or admit them if the required threshold is achieved based on the additional facial features. Secondly, the semantic analysis and precise (objective) facial measurements permits the effective use of extensible indexing or other forms of database filtering for high-speed searching in identification mode.
|
