Transcript Slide 1
Face Recognition Using Eigenfaces Obama and Biden, McCain and Palin Justin Li What’s Face Recognition Good For? Security Systems Smart Artificial Systems New Human-Machine Interface Methods A Survey of Methods Eigenfaces Facial Measurements Mapping 3D Morphable Model Facial Feature Mapping Brief History of Eigenfaces Matthew Turk Alex Pentland Michael Kirby Lawrence Sirovich The Eigenface Approach (1) • Name comes from the use of eigenvectors to identify faces. • Principal Component Analysis – Takes the mean of the pictures from a grayscale training set. – Subtracts the calculated mean from each picture. – Forms a covariance matrix. – Finds the eigenvectors for the covariance vectors. • PCA gives the resultant eigenvectors/eigenfaces. The Eigenface Approach (2) • Weighs test images with the eigenvectors to find correlation. • A computationally fast method for face recognition. • Note that the covariance matrix will be exceedingly large. – A simplification is introduced. – Instead, make a simplified matrix with dimensions the number of pictures in the training set. – Scale the eigenvectors using the training pictures. • This allows for a computationally feasible way to calculate the eigenfaces. Flaws and Limitations • Lighting conditions • Requires specific alignment and normalization. • Face orientation Implementation and Improvements • Standard implementation as given in the paper by Pentland and Turk. • Improvements possible with better segmentation and cropping, perhaps ignoring more portions of the hair. • Combining method with results from other facial metrics or from other methods entirely. Demonstration and Questions?