Chinese researchers take top honors in cartoon face recognition challenge

Categories Biometric R&D  |  Biometrics News

A Chinese entertainment company last week announced the winners of its challenge to write biometric algorithm models that can detect and, separately, recognize cartoon character faces.

IQIYI Inc., which hosts cartoon and live-action entertainment on its Chinese language-only site, said that it held the iCartoonFace Challenge to research recognition techniques for use in entertainment. All major participants were from China.

Researchers from Southeast University in Nanjing finished first in character detection, with a mean average precision of .9291.

Teams from Zhejiang University in Hangzhou and Sun Yat-sen University in Guangzhou jointly achieved a top-ranked 92.4697 percent accuracy rate for character recognition.

They were among 481 groups, from both university and technology firms, participating in the challenge. Companies assigning teams included JD.com, Baidu, Tencent, Alibaba and SenseTime.

IQIYI executives said that they had assembled the world’s “largest known manually tagged datasets” for cartoon recognition and detection training. More than 5,000 characters were captured in the 400,000-image set.

Executives with iQIYI had announced in April what they said would be the first “large-scale cartoon character recognition” contest in China. They said it takes better software to detect and recognize a drawn or animated face than the real thing.

It might seem counterintuitive that the stuff of Saturday morning television programming in the United States — cartoon characters — are more complex than humans in any way. Yet, cartoon characters are the creations of artists’ unique imaginations. By contract, all human faces in all their diversity follow shared biological design rules.

Precise algorithms can create mash-up video content, according to the company, based on themes to create new content.

IQIYI has hosted two video multimodal biometric recognition challenges over the past two years. They shared the same goal — raising the accuracy of algorithms.

Related Posts

Article Topics

 |   |   |   |   |   |   |   |   |   | 

Comments

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Most Read This Week

Featured Company

Biometrics Research

Biometrics White Papers

Biometrics Events

Explaining Biometrics