View Full Document

Memory-based Face Recognition for Visitor Identification



View the full content.
View Full Document
View Full Document

27 views

Unformatted text preview:

Appears in Proceedings of International Conference on Automatic Face and Gesture Recognition 2000 Memory based Face Recognition for Visitor Identification Terence Sim1 Rahul Sukthankar1 2 1 The Robotics Institute Carnegie Mellon Univ Pittsburgh PA 15213 Matthew Mullin2 Shumeet Baluja1 2 2 Just Research 4616 Henry Street Pittsburgh PA 15213 tsim rahuls mdm baluja justresearch com Abstract We show that a simple memory based technique for appearance based face recognition motivated by the realworld task of visitor identification can outperform more sophisticated algorithms that use Principal Components Analysis PCA and neural networks This technique is closely related to correlation templates however we show that the use of novel similarity measures greatly improves performance We also show that augmenting the memory base with additional synthetic face images results in further improvements in performance Results of extensive empirical testing on two standard face recognition datasets are presented and direct comparisons with published work show that our algorithm achieves comparable or superior results Our system is incorporated into an automated visitor identification system that has been operating successfully in an outdoor environment since January 1999 1 Introduction The problem of visitor identification consists of the following a security camera monitors the front door of a building acquiring images of people as they enter an automated system extracts faces from these images and quickly identifies them using a database of known individuals The system must easily adapt as people are added or removed from its database and the system must be able to recognize individuals in near frontal photographs This paper focuses on the face recognition technology that is required to address this real world task Face recognition has been actively studied 7 12 particularly over the last few years 9 The research effort has focused on the subproblem of frontal face recognition with limited variance in illumination and facial expression In this domain techniques based on Principal Components Analysis PCA 10 popularly termed eigenfaces 26 16 have demonstrated excellent performance This paper introduces a simple memory based algorithm for face recognition termed ARENA that satisfies the requirements outlined above and also significantly outperforms PCA based methods on two standard face recognition datasets 2 Image Datasets and Preprocessing Our results use human face images from two standard datasets Olivetti Oracle Research Lab ORL 22 and FERET 17 19 ORL consists of 400 frontal faces 10 tightly cropped images of 40 individuals with only minor variations in pose 20 illumination and facial expression The faces are consistently positioned in the image frame and very little background is visible FERET contains over 1100 faces however many of them are unsuitable for our experiments since they are partial or full profiles or the individuals were only photographed twice Therefore from FERET we selected the subset of images that satisfied the following two constraints 1 near frontal poses 2 images of individuals with more than five such images our tests require several images for each person The resulting 275 images consist of 40 individuals with greater variation in pose and lighting than in the ORL dataset For instance many of these images were taken over different days and display significant differences in hairstyles eyewear and illumination Unlike the ORL images the FERET faces are of non uniform size and do not always appear in the same location of the image We use the FERET images as provided to explore the potential limitations of our templatebased face recognition technique Figure 1 shows two images for each of two individuals from the two datasets Figure 1 Top row Two sample images each of two subjects from ORL left and FERET right Note the difference in facial orientation expression and accessories between the two images of the same individual FERET images tend to exhibit greater variation in appearance including hairstyle and clothing Bottom row the corresponding ARENA reduced resolution images 16 16 pixels 3 The ARENA Face Recognition Algorithm ARENA is a memory based 1 algorithm that employs reduced resolution images 16 16 and the L 0 similarity measure described below The reduced resolution images are created by simply averaging over non overlapping rectangular regions in the image The distance from the query image to each of the stored images in the database is computed and the label of the best match is returned 3 1 L p similarity measures Our results show that the obvious choice for ARENA s similarity measure the Euclidian distance performs poorly In this section we present alternatives The Lp norm is de1 P fined as Lp a ai p p Thus the Euclidian distance is simply L2 x y Note that since we are not interested in the actual distances but only in the ordering we P can equivalently employ the similarity measure L p a ai p Robust statistics literature shows that L 2 despite its convenient analytic properties overly penalizes outliers 11 For this reason the L 1 similarity measure is often used in noisy environments For ARENA we have explored several L p similarity measures see Figure 2 The L 0 similarity measure is defined as L 0 a limp 0 L p a Intuitively L 0 x y counts the number of components in x and y that differ in value Our experiments indicate that the best performance on this task is achieved with p 1 In our application each reduced resolution image is converted into a vector x where each pixel in the image is represented as a component of the vector In practice since individual pixel intensities are noisy we relax the definition of L 0 to be X L 0 x y 1 xi yi where is a threshold such that pixels whose intensities differ by less than are considered equivalent 4 Principal Components Analysis PCA The most widely used baseline for face recognition eigenfaces 26 16 employs Principal Components Analysis PCA which is based on the discrete Karhunen Loe ve K L or Hotelling Transform 10 is the optimal linear method for reducing redundancy in the least mean squared reconstruction error sense Points in Rd are projected into Rm where m d and typically m d PCA has become popular for face recognition with the success of eigenfaces 26 For face recognition given a dataset of N training images full resolution originals each with d pixels we create N d dimensional vectors x1 x2 xN where each


Access the best Study Guides, Lecture Notes and Practice Exams

Loading Unlocking...
Login

Join to view Memory-based Face Recognition for Visitor Identification and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Memory-based Face Recognition for Visitor Identification and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?