09/02/2026 6:48 PM - PACES
09/02/2026 6:48 PM - PACES
Aidan Dempster - adempst - 6125 9596
Problem:
How can we efficiently recognizes faces in images in a way that allows the easy addition of new faces?
Approach:
Instead of using standard features of the face like the eyes, ears, or nose, they learn the 2D components of a face that explain the maximum variance in the set of images of faces. They argue that representing an image in a "face space" made of the top M highest variance eigenfaces, they construct an information dense feature space to allow the recognition of faces.
Contribution:
They propose to automatically learn what stimulus is important for facial recognition by applying PCA to a dataset of faces. They proceed to define procedures for locating faces in images and performing facial recognition by applying their "face space".
Evaluation:
They construct a dataset of faces with variations in head size, head orientation, and lighting conditions, and evaluate the performance of the method in a recognition task.
Substantiation:
The evaluation gives an idea of how sensitive the method is to variation in problem setup, but does not substantiate the claimed increase in performance over other methods.