The novelty of the project, also reported in a recent issue of IEEE Transactions on Multimedia, do2

Local Antioquia Medellin Judicial Education Infrastructure Public Order National Health Agriculture Infrastructure Bogota Judicial Law Enforcement Education Health Economics Economic Indicators Currency Converter Medellin Policy National Peace Process Presidency Elections Congress 2014 International North America Latin America Africa Asia Europe Oceania Sports League Postobón League Results 2014 World Cup Postobón Athletics Cycling Motor Sports International Football do217 Golf Tennis Other Sports Entertainment
Researchers at the National University of Singapore have developed a computer program that allows to accurately estimate the age of a person. To achieve this, scientists have collected hundreds of pictures and videos of faces taken from the Internet, comprising the largest database of its kind created so far.
The aim is that the application acquires a universal character, for snapshots including people of different racial, taken with different do217 lighting do217 conditions and image quality.
Although for humans can be almost a game to predict do217 the age of the other person just take a look at his face, for computers is a much more complicated, time-consuming do217 and painstaking work. There have been several studies on this matter, and the latter aims to achieve universal estimator of human age.
As reported in the science and technology portal do217 Physorg, the researcher Bingbing Ni Advanced Digital Sciences Center, along with Zheng Song and Shuicheng Yan from the National University of Singapore, do217 has developed a system to accurately estimate the age of a person to through facial recognition, regardless of their ethnicity or quality and lighting of the picture.
The novelty of the project, also reported in a recent issue of IEEE Transactions on Multimedia, do217 American academic journal on Advances in multimedia technology and its applications, is the creation of a large database with which to train the system before testing it in a real way.
To do this, the researchers developed an automatic web image and video mining project, which involves the extraction of previously unknown information that could be useful, in which they used search queries related to age to collect nearly 400,000 images the most popular search engines, such as Flickr and Google, as well as 10,000 YouTube videos.
On one hand, the images of faces which were labeled with age so that they could be used to develop a new learning algorithm with which prepare the system estimation. For the search, we used human faces multiple detectors based on different existing techniques. Images with high similarities were grouped together, resulting in a database with about 175 faces per group.
On the other hand, researchers downloaded videos, of which sequences extracted face tracking. Although in this case is not labeled with the years, the face tracking within the same sequence was carried people from the same age group, restraining representing an additional regularizer to increase system efficiency.
The result, after removing the images of poor quality and wrong, was a database with 77,000 snapshots, which in turn contain 219.000 human faces. Furthermore, unlike some previous databases smaller do217 faces this includes people from various racial, lighting conditions and heterogeneous do217 quality.
Another difference with previous programs is that by using the web-based data mining, "do not need to manually do217 build a database with images of faces, do217 which is rather laborious and costly," said Prof. Ni.
By contrast, this estimating system has the ability to generalize, so that the shared in the database during the training process can be applied to faces generally once really works.
With a standard image quality, the system is able to calculate the age of a person with a maximum margin of error five. In the future, researchers hope to improve even more this software tool, minimizing errors, especially with blurry images or low resolution deteriorated.
Already they have sought a wide variety of applications, such as demographic data collection in supermarkets and other public areas, age-specific human-computer interaction or identifying people through do217 old photographs