Researchers from the University of the Basque Country have applied computer vision and machine learning technology to predict facial beauty based on images and better calculate the age of the people depicted.
Artificial intelligence research covers many fields, and aims to realize the ideal intelligent machine to perceive its environment and perform actions that maximize the chance of success for certain goals or tasks.
In this case, the team Computer Vision and Pattern Discovery Basque National University (UPV/EHU) research applies computer vision and machine learning to biomedical images (detecting cells, tissues, tumors…), streets (positions of vehicles, pedestrians…) and even facial image pictures, To better perceive beauty and estimate age, just like in the study just published in the journal Expert system with applications.
“Basically, we use modern artificial intelligence technology to solve a variety of image problems, including 2D, 3D, video and other images.” UPV Computer Science and Artificial Intelligence Department Ikerbasque researcher Ignacio Arganda explained. / EHU.
He added: “They are usually machine learning techniques, because usually we start with a set of labeled data, images or videos (knowing where the objects are or what type of category they have), and we use this to teach or Train our statistical or artificial intelligence models to assign the same labels to examples they have never seen before”.
“When predicting aesthetics, we try to use semi-supervised techniques (in which not all images are labeled) to replicate the aesthetic scores given in different databases. –Explaining Arganda–. To this end, we use a network that has extracted different features to train the model to predict beauty.”
Along these lines, team members have shown that semi-supervised learning, which has never been used for this type of problem before, provides even better results than supervised learning (in which all images are labeled).
“Regarding the estimation of age, a convolutional neural network (CNN) is used: what we have is the input image; the researchers pointed out that a series of filters will extract features that help make the final decision, that is, a number. In the case of “age”.
Researcher Ignacio Arganda. / UPV / EHU
The author conducted an empirical study to see which error functions can help train the network in this area better, because the estimation error can be minimized in different ways. Experimental results demonstrate how to improve age estimation.
Deep neural network
For this type of prediction and estimation, deep neural networks are used: “They are networks with many connections, many filters, and millions of data… But it’s important to understand the language used by the network to predict a person’s age , Or perform any other type of prediction. Currently, there is another area of research that we are immersing in, the so-called interpretable or interpretable artificial intelligence. This research aims to clarify the technology that the network focuses on”, Arganda explained .
Similarly, researchers warn that the impact of machine learning technology on our lives is beyond our imagination. “A lot of data is being generated, and high-level decisions are being made based on these systems. It is very important to consider ethical factors.”
In fact, in machine learning, a huge database is used to train the model, and all the deviations contained in the data are replicated in the predictions and estimates made by the model, which is indeed harmful. Currently, there are public studies in which how to remove different biases from the data.
Fadi Dornaika, Wang Kunwei, Ignacio Arganda-Carreras, Anne Elorza, Abdul Abdelmalik Moujahid [y S.E. Bekhouche]. “Towards a graph-based semi-supervised face beauty prediction”. Expert system with applications (2019). DOI:10.1016 / j.eswa.2019.112990