Facial processing in realistic scenarios

Facial biometrics technologies are directly useful in the field of security: access controls to facilities, systems, video surveillance and biometric computer passwords are some of the most characteristic areas in which Gradiant carry out intensive research both theoretical and of application

Facial biometrics technologies are directly useful in the field of security: access controls to facilities, systems, video surveillance and biometric computer passwords are some of the most characteristic areas in which Gradiant carry out intensive research both theoretical and of application. The ultimate goal is to improve the systems performance under adverse conditions, allowing the deployment of this technology in complex and realistic scenarios. Undoubtedly, this is the current trend in face recognition, where the prototypes thoroughly tested under controlled or semi-controlled come up against images of the “real world”. The Projects Labeled Faces in the Wild (http://vis-www.cs.umass.edu/lfw/) and Public Figures Face Database (http://www.cs.columbia.edu/CAVE/databases/pubfig/) are two good examples, with thousands of images collected directly from the Web via semi-automatic tools and protocols that allow comparison of different algorithms in these complex scenarios.


Gradiant have conducted at a theoretical level, research on statistical modeling of the coefficients resulting from a Gabor multi-orientation and multi-scale filter, works that have been published in various international conferences, highlighting those published in the IEEE International Conference on Image Processing (ICIP) in the years 2007 and 2009. These investigations have led to a drastic reduction in the size of biometric models (templates) while maintaining the performance of the original system, enabling the use of chips of low storage capacity (eg RFID tags) to hold biometric templates. In the same line, a project carried out in collaboration with the Department of Signal Theory at the University of Vigo was accepted late at the end of last year for publication in the next IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) 2010, to be held in Dallas, Texas (USA).
The choice of appropriate statistical models has also allowed taking the first steps towards the selection of optimal distances compared biometric systems based on a Gabor filter, topic which has not aroused much interest among the international scientific community and undoubtedly will improve the performance of such systems.

Not everything was going to be security…
In addition to security applications, the automated systems for detection and analysis of faces have also been directed towards other areas where i factors such as accessibility, usability and personalization are priority. Thus the worlds of man-machine interfaces and video games have recently been revolutionized by these and other technologies. No more than a glance at the commercials PlayStation (EyePet) and Xbox (Project Natal)is needed to verify the importance they have for these projects, the automatic face recognition, gestural and emotional modules, the capture of the user’s body movement, the voice recognition and the use of “augmented reality (augmented reality, AR) to create virtual pets in a life-like environment.

In Gradiant there exist lines of research linked to many of the technologies just listed: facial analysis systems can automatically detect and recognize the user in front of the camera, and also provide demographic information (sex, age range, race) and emotional for a greater degree of customization and interaction with the machine. Combined with the use of 3D models (avatars), you can transfer the avatar gestures and expressions to create a virtual and dynamic representation of user within the game. Likewise, systems which use face and facial features detection allow controlling the PC cursor just by a head (or eyes) movement and “diving” into computer-generated 3D scenes, which could be those of an action game in first-person.

Within this line last year, Gradiant developed a project for subjective immersion in three-dimensional computer-generated images, work that was awarded the third prize in the contest IX of Expourense Tecnolóxico Proxectos. Another project currently open contemplates the selective masking of faces (e.g. children) in video sequences in order to preserve the privacy of minors.

Gradiant is currently involved in a project funded under the Avanza Plan call 2009 R & D aimed at developing an interactive and personalized platform. This platform capable of being placed in a mall or shop windows, is of great interest to the trader as it allows obtaining objective measures of success of a particular advertising campaign: number of people who have shown an interest, hours of business, etc. . . . Classification systems allow a more granular study, as the data mentioned above can be broken down into sectors of the population. From the the potential client’s point of view the advertising platform offers “customized” products based on the knowledge of sex and age of the subject in front of the screen.

For its part, interactivity is another strong point of the project, it’s very important in order capture the attention of potential customers. In this section, the processing of faces has a key role by allowing, among other options, the virtual test of glasses and communication with computer-generated avatars.
The main lines of action within the above projects go through a refinement of facial processing systems which allow its use in scenarios of high complexity. On the other hand the Multimodal Information Specialist Group in Gradiant, are still looking for novel applications where the use of this technology has a positive impact on society or involves an improvement over currently available solutions.

Share on FacebookGoogle+Tweet about this on TwitterShare on LinkedIn