Face Detection Software Used to Determine Female-to-Male Ratio at BarsFace Detection Software Used to Determine Female-to-Male Ratio at Bars

An Austin startup, SceneTap, has developed an application that determines vital stats for a bar like the male to female ratio, how full the bar, and the average age of the men and women in the bar. This is done using face detection technology that picks up characteristics to estimate the age and gender of the person. Then the data is available through an smart phone app for the bars you are looking to go to.Right now it is only available at some San Francisco bars, but it is interesting to me to see face recognition technology migrate from security  (is this person a terrorist?) to nightlife (which bars are hopping?) applications.

This data is useful for the consumer wanting to know if the the bar is full and what kind of crowd is at the bar, and for the bar management to know if their marketing is attracting the type of crowds they want. You can read more about it at SF weekly.

Face Recognition Software that can Scan Through Millions of Faces Per Second

I received a couple of emails today about a face recognition system from Hitachi Kokusai Electric that can search up to 36 million faces per second. One of the impressive parts is that it can recognize a face with 30 degree deviation from the camera and the small image size required 40 x 40 pixels. Like almost all commercial face recognition systems, this requires a man in the loop to verify the images. The article doesn't give the accuracy, but usually for a face recognition system to be useful there needs to be an accuracy high enough for a person verifying the images to not be overwhelmed with images of suspects.

Errors in face recognition are generally caused by three factors: pose errors i.e. the person having a there head in a different position than the stored image or training image of the person, lighting issues i.e. the lighting on the person is different than the training image, and a catch all called non-rigid transforms. Non-rigid transforms includes most changes in a person's appearance from the stored image of the person to image one is trying to match to the image. These include changes in skin color like a person getting a tan from a day at the beach or have pale skin because they are sick, as well as, occlusions to face like sunglasses, scarfs, and the new phenomenon of face recognition dazzle. I talk about these errors more in an article I wrote about face authentication and in my PhD dissertation.

I'm a little dubious that this software will be as effective as it claims. While fixing pose and lighting errors are almost solved problems by using 3D modeling of heads to artificially replicate the pose and lighting, the non-rigid transforms still pose a problem. So, this system may not work if you get a tan or put on sunglasses. 

What is Pattern Recognition?

Pattern recognition is, in a nutshell, grouping, description, recognition, and classification of perceptual structure. In other words, I have some data (images, sounds, inputs, etc.), what information can I get out of it?

Dr. Hamed Sari-Sarraf, my image processing professor at Texas Tech, described pattern recognition and image processing as the two sub-disciplines that make up computer vision. He said, that image processing converts images into data that is useful e.g. remove noise, highlight regions of interest, etc.; whereas, pattern recognition is using those processed images to make a judgment about the data e.g. identify a face, perform a classification of an object, etc.

According to a former co-worker, Don Waagen, pattern recognition is broken down into four basic parts from lowest level processing to highest: sensing, segmentation, feature extraction, and classification. Sensing converts images, sounds, x-rays, etc. to a signal. Segmentation isolates sensed objects from uninteresting signal i.e. noise. Feature extraction measures useful properties of objects e.g. width of a face image or the length of a bridge; basically, a feature is any information from an object that can be unique to that object. Classification assigns objects to a category.

Standing on the shoulders of Hamed and Don, I view pattern recognition as feature extraction and classification of data. I consider sensing and segmentation as more image processing. But, like anything, the lines between the topics are blurry, and I will discuss some image processing topics; however this blog will mainly focus on pattern recognition.

About Me

Greg Wagner, PhD

My name is Greg Wagner. I work as a algorithm developer focusing on pattern recognition and computer vision problems. I have a PhD and MS in computer science from Texas Tech University and a BS in computer science from New Mexico State University.

I have a few articles on automatic target recognition (ATR) ("Error estimation procedure for large dimensionality data with small sample sizes" and "Analysis of uncompensated phase error on automatic target recognition performance") and face authentication ("Cascading trilinear tensors for face authentication"). My dissertation was on face authentication in which I feel I have a good description of SVMs. I have not published much since I graduated, but I have spoken about ATR and face authentication at my internal company conferences and at a few external conferences; as well as, I have taught company continuing education classes on ATR, image processing and pattern recognition.

This blog will contain posts that talk about pattern recognition topics as well as links and commentary to articles I find interesting.