face detection

Henry A. Rowley, Shumeet Baluja, Takeo Kanade

---

(easy cases, see paper 1 below)

---

(difficult cases, see paper 2 below)
(The eye locations are merely to show the orientation of the detected face. They were not located by the detector.)

---

Thesis Dissertation

You can download my thesis dissertation [9.5 MB gzipped postscript, about 30 MB uncompressed].

Demo and Results

We digitized some video sequences at about 3-5 frames per second at a resolution of 320x240 pixels. These images were then processed by the face detector, taking about 5-10 seconds per frame on a 200 MHz SGI Indigo. The results were then MPEG encoded, resulting in the following four videos: [person walking] [web browsing] [person walking again] [crowd]

Papers

1. , by Henry A. Rowley, Shumeet Baluja, and Takeo Kanade. IEEE Transactions on Pattern Analysis and Machine Intelligence, volume 20, number 1, pages 23-38, January 1998. Neural Network-Based Face Detection
2. , by Henry A. Rowley, Shumeet Baluja, and Takeo Kanade. Accepted for oral presentation at Computer Vision and Pattern Recognition, 1998, pages 38-44. Also available as CMU CS Technical Report CMU-CS-97-201, 1997. Rotation Invariant Neural Network-Based Face Detection

You can also look at Henry Rowley's thesis proposal document and proposal presentation slides, 1996. Also available are some older papers whose content is covered by the papers above.

Test Data

The test images used in the above papers are available. I encourage other people working on face detection to use these images, to provide a standard for comparison.

Source Code Download

It is free, but you use the software at your own risk.

You need input your your name, email and organizartion from the following download page, or send an email to to Dr. Changbo Hu if you have difficulties in downloading.

Source Code Download

Applications

Several researchers have incorporated this face detection system into their own projects:

• Michael Smith at CMU has used the detector as one feature for generating summaries, or "skims", of video sequences.
• Shin'ichi Satoh has used the detector as part of a system called Name-It, which automatically associates names with faces in video.
• WebSeer was a search engine for images in the web, developed at the University of Chicago. It used many features for matching images to your query, including the results of our face detector for locating portraits.

Contact Information

For more information, please contact a current member of the Face Group.

Related Links

Below are a few resources related to facial image processing on the web. Please let me know if there important items missing.

• This work was part of the Face Group at CMU.
• This work is considered part of the MURI project at CMU.
• Face Recognition Home Page (this page doesn't seem to exist any more, if you know the new location please let me know...
• CMU's project on facial expression analysis.
• Henry Schneiderman (CMU) is doing research on face detection using probabilistic models.
• Real-time face detection project (NASA) based on color features.
• Beifang Qiu's (Vanderbilt) work on face detection uses a variety of techniques.
• Leiden University's work on face detection based on information theory.
• Thomas Leung's (Berkeley) work on face detection using random graph matching.
• Shang-Hung Lin's (Princeton) work on face detection and recognition using neural networks.
• MIT Media Lab's work on face detection and recognition using eigenfaces and related probabilistic methods.
• MIT AI and CBCL Lab projects on object detection using neural networks, clustering, and support vector machines.
• Demonstration of the "Thatcher Illusion".

---

Henry A. Rowley (har@cs.cmu.edu)

Shumeet Baluja (baluja@cs.cmu.edu)

Takeo Kanade (tk@cs.cmu.edu)