12px13px15px17px
Date:13/05/17

Researchers have created the most accurate digital model of a human face

A team of researchers from Imperial College London have created the most accurate to date digital model of a human face. Their development, according to the creators, «the maximum average features», and so it can be used to create three-dimensional models of human faces in video games, movies, systems and technologies to improve face recognition.
 
As a rule, when you create computer models of individuals uses the method of three-dimensional morphological model (3D morphable model, or 3DMM). As «the basics» used something like a frame, which is a set of parameters characterizing the deviation of key points. Each parameter must be configured manually, and then bring to mind the final model in a 3D editor. British scientists have developed a method to fully automate the process of compiling a 3DMM model that allows to include data obtained from a sufficiently large number of people. Their algorithm automatically builds snapshots of one person's face according to their spatial orientation and their base is a three-dimensional model. In addition, using the same algorithm to detect errors and remove them from the final results.
 
In the process of developing a computer system has analyzed a huge number of images of faces of people of different age, nationality and gender. The data was compiled medium model of a human face, called facial Large scale model (LSFM). New LSFM model was used to generate 100's of thousands of images used to train the artificial intelligence based on neural network. Then the neural is able to convert two-dimensional images of people's faces in a fairly accurate three-dimensional models.
 
The researchers are now trying to improve the algorithm in order «to teach» a system to recognize human emotions, which will significantly improve the already rich possibilities of the new method of 3D-modeling.




Views: 309

©ictnews.az. All rights reserved.

Facebook Google Favorites.Live BobrDobr Delicious Twitter Propeller Diigo Yahoo Memori MoeMesto






21 November 2024

20 11 2024