Demonstration of NEC soft's FieldAnalyst: The software estimated a reporter from Tech-On! (28-year old female) to be a male in his 20's.
Demonstration of NEC soft's FieldAnalyst: The software estimated a reporter from Tech-On! (28-year old female) to be a male in his 20's.
[Click to enlarge image]

The use of face recognition functionality is gradually increasing in digital cameras, mobile phones, etc. A number of demonstrations using this function were performed at CEATEC Japan 2007, which ran from Oct. 2 to 6, 2007 at Makuhari Messe.

NEC Soft Ltd. and Mitsubishi Electric Corp. both gave demonstrations of how this function can be utilized on the streets. At NEC Corp.'s booth, NEC Soft exhibited FieldAnalyst, personal computer software for estimating sex and age groups, which was launched on Oct. 2, 2007. It is expected to be used in commercial facilities such as supermarkets and exhibition halls.

The software recognizes a face based on an image obtained with a camera and estimates sex and age group. The target applications include marketing surveys to analyze customers and advertising content that changes in response to each customer.

The software quantifies facial characteristics and checks thus obtained value against the data stored in the database, thereby estimating sex and the age group based on the closest value in the database.

According to the company, the estimation accuracy rates of sex and age group are 90% or higher and about 60%, respectively. The accuracy rate of the age group estimation is lower because the ages are divided into groups such as 10 to 19 and 20 to 29. Practically, the accuracy is almost on par with that of human judge.

Mitsubishi Electric incorporated a function to estimate sex and the number of people into "gate-shape multi DLP," a system composed of multiple DLP rear projectors installed in the shape of a gate, and exhibited the setup.

For example, it is expected to be used in an application where the gate-shape multi DLP is installed at the entrance of commercial facilities to display the information on the shop determined to be the most appropriate based on sex and the number of people who walk through it. Specifically, the system recommends a bar if a couple walks through it, while introducing a cafe if the passerby is a female.

Because the system has to provide information to passersby, the recognition speed is most crucial. In the demonstration, the system required 1/3 second for the face recognition. While the accuracy of recognition is important to a certain degree, in an actual application, "it is the way of outputting recognition results that matters," said a demonstrator from Mitsubishi Electric.

In case of a false recognition, a female customer would be offended if the system indicated "you are a male." No serious problems occur, however, if the system displays information on a restaurant preferred by both male and female. The company reportedly completed the technical development including the face recognition and further development toward commercialization, such as the evaluation of content, is ready to be initiated at the request of clients.