190 likes | 217 Views
Explore the world of biometrics and the significance of face recognition in authentication. Learn about different face recognition techniques such as Eigenface and Neural Networks, along with their algorithms and applications. Discover the challenges faced in face recognition and innovative solutions to enhance accuracy and efficiency.
E N D
BIOMETRICS Bhavin Shah CIS 203 12/04/2003
What is Biometrics • Measurable physiological and/or behavioral characteristics that can be utilized to verify the identity of an individual • Mainly include fingerprints, retinal and iris scanning, hand geometry, voice patterns, facial recognition, facial thermogram, ear geometry
Need for Biometrics • Growing interest in biometric authentication • Photo-ID cards, Airport Security, Office Surveillance • After 9/11, extensive security problems • Required at several highly secure and military establishments
Why Face?? • Most recognizable part of human body. • Unique for every person (exceptions: identical twins, but they can be distinguished using one of the techniques discussed later on) • Easy to acquire • Can be used even without the knowledge of the subject
Some Terms • Face Enrolment • Associating a face in a given image with a given label (subject’s name) • Face Identification • Given a face-image and a gallery of class-labeled faces, finding the correct face. • Face Verification • Verifying that a given label is associated with the face in the given image
Face Recognition Techniques 2D Face Recognition • Eigenface • Distinctive characteristics are highlighted • Feature Analysis • Localization of features • Neural Networks • Back propagation techniques • Graph Mapping • Construct a graph around the face • Automatic Face Processing • Uses distances and distance ratios • Fisherface • Uses ‘within-class’ information to maximize class separation • Etc…
EIGENFACE Patented by MIT Utilizes two dimentional, global grayscale images Face is mapped to numbers Create an image subspace(face space) which best discriminated between faces Can be used in properly lit and only frontal images NEURAL NETWORK Employs an algorithm to determine similarity between enrolled and verification face Assigns a symmetry magnitude to each pixel, to create a symmetry map Several neural-networks are trained using various back-propagation methods Comparison
ALGORITHM The Training Algorithm • Gabor filter face image • Apply Gaussian weighting • Locate peaks in image • Extract feature vector at located peaks • Store vector, location and class label for each of the located peaks
ALGORITHM Gaussian Weighting Gabor Filter Original Image Gabor Filtered Image 2-D Gaussian Image Gabor * Gauss Locate feature points
ALGORITHM The Testing Algotithm • Step 1-4 are same as Training algorithm • For each extracted feature vector, compute distance to all feature vectors in the gallery • Based on class label to the nearest matching feature vectors, assign points to corresponding class
Disadvantages of Face Recognition • Not as accurate as other biometrics • Good quality and well-lit images are needed • Large amounts of storage needed • Image-shot important (frontal, side, back,etc) • Different facial expressions will lead to different image mapping
SOLUTIONS (1) LIGHTING • Controlled lighting • Dominant light source • Filters • Edge detection • Color normalization • Brightness and hue
SOLUTIONS (2) IMAGE-SHOT (Pose) • Frontal image required • Subject image should not be obstructed by any third object • Multiple images with different facial expressions can be stored
Modeled according to skin color Original Image Image after color normalization
Image after filtering out clutter Identified Face Image
MY IDEA!!! • Capture the image of the subject using 3 different cameras • Frontal image, right-side image and left-side image. • Store in 3-D coordinates • Take multiple images with different facial expressions
MY IDEA!!! • Not to look at the whole picture, rather just look at the frame where there is a lot of skin color. • This would speed up the algorithm and also eliminate false detections. • Slightly Longer and time consuming idea: • Detect corners at every part of the face, eg. Eyes, ears, nose, chin, etc… • Match the corner with the corner saved in the database at the time of enrolment.
Some Testing!!! • Face recognition Link • http://www.ifp.uiuc.edu/~antonio/Demo/SelectImage.html