A3KellerClark

A3KellerClark

To Do

• please work as a group (minimum: 2 members; max: 6 members) and submit one answer as a group (clearly identifying the members of your group)
• identify one focused topic within the chapter which is of greatest interest to your group!
• each group member: should identify one additional source relevant to the topic chosen!
• each group: provide a two page max summary statement in our course environment (mention the additional resources identified)
• prepare a short presentation to the class for the class meeting on Sept 17! the groups can choose how to present their results (oral only; use slides; one member, several members, or all members); time allocation (will be dependent of the numbers of groups: but somewhere between 4 and 10 minutes)

Form for your response

1. Herbert Simon

Scott Keller, Matthew Clark

2. Most interesting idea/concept you learned from the article?

One topic in the chapter deals with automatic photo recognition. As digital cameras make it easier and easier to build up huge collections of images, it becomes harder and harder to organize them. Even a well-kept structure of directories and subdirectories with well-named images is difficult to navigate. Also, most users are more likely to create a single directory of images with names like "DCS0032.JPG." Projects like the Google Image Labeler, http://images.google.com/imagelabeler/ improve image search by having humans tag images. Other approaches are aimed at having computers learn to process images directly.

The Carnegie Mellon Robotics Institute has several projects working on completely automated face recognition, face detection, facial feature detection, and expression analysis.

http://www.ri.cmu.edu/labs/lab_51.html#projects

PIE database:

The PIE database is a database of pictures of people's faces, ranging across Pose, Illumination, and Expression. CMU uses this database to test facial recognition projects.

Between October and December 2000 we collected a database of 41,368 images of 68 people. By extending the CMU 3D Room we were able to image each person under 13 different poses, 43 different illumination conditions, and with 4 different expressions. We call this database the CMU Pose, Illumination, and Expression (PIE) database.

http://www.ri.cmu.edu/projects/project_418.html

Light-Fields for Facial Recognition:

The most important decision in developing an object recognition algorithm is selecting the features on which to base the algorithm. The most common approach is the appearance approach, which uses the pixel intensity values of an object as the features used to recognize it. Pixel intensity is caused by reflected light from the object along a particular angle. Pixel intensity changes when the object is viewed from a different angle, or in different lighting. A new approach is to estimate the "light-field" of an object, as the set of all of the angles of view, in all possible lighting conditions.

http://www.ri.cmu.edu/pubs/pub_4058.html#abstract

Feature Recognition

After an image is scanned, an algorithm analyzes the light-field or pixel intensity and plots key landmarks on the face. Landmarks are translated into nodes which mark features on the face such as eyes and nose. The distances are measured then between these nodes and matched to a database to find a match.

W.Y. Zhao, R. Chellappa, Image-based Face Recognition: Issues and Methods, Image Recognition and Classification , Ed. B. Javidi, M. Dekker, 2002, pp. 375-402

Surface Texture Analysis

To increase positive facial and feature matches, STA is used. A high-resolution image is taken of a small portion of a persons skin. This "skinprint" is then mathematically reduced to blocks and measured in the same way features are. The result is the ability of the system to distinguish any lines and pores on the face and to determine the persons skin texture. This has been shone to increase positive matches by about 25% and has given facial recognition software the ability to identify identical twins.

-- http://www-users.cs.york.ac.uk/~nep/tomh/Biometrics.html http://securitysolutions.com/mag/security_biometrics_mainstream/

3. articulate what you did not understand in the article but it sounded interesting and you would like to know more about it

R. Gross, I. Matthews, and S. Baker, Appearance Based Face Recognition and Light-Fields, tech. report CMU-RI-TR-02-20, Robotics Institute, Carnegie Mellon University, August, 2002.

chosen by Matthew Clark

Woodward, John D, et al. "Biometrics: A Look at Facial Recognition." RAND Public Safety and Justice. 2003. http://www.rand.org/pubs/documented_briefings/DB396/DB396.pdf

chosen by Scott Keller