The CCTV program Waiting For Me made a special report on May 3 that the police finally found a missing child who had been abducted over 10 years with the help of "cross-age facial recognition" technology provided by Chinese tech giant Tencent. 

It was reported that China has a number of missing persons every day. Due to rapid population mobility, it is becoming increasingly difficult to search for missing persons as time goes by. Efforts looking for missing persons have been described as "looking for a needle in the ocean".  The facial recognition technology developed by Tencent has helped solve this problem.  

A demonstration of facial recognition technology developed by Tencent. [Photo provided to chinadaily.com.cn] 

Based on its years of experiences in the field of computer vision, Tencent Youtu Lab has been committed to improving its facial retrieval technology to help police find missing persons by comparing, locking, and matching faces in its massive facial database. The accuracy of Tencent's massive data retrieval has reached 99.99 percent and it can retrieve tens of millions faces within a millisecond, saving the police manpower and material resources. Police in Fujian and Sichuan have employed the technology to search for missing persons and with the help of the technology, the Department of Public Security of Fujian Province has brought 1,091 people back to their families.  

In previous cases, most of the missing persons could be only retrieved within the short period of time after they were first reported missing. To help Sichuan police deal with its long-pending cases, Tencent needed to overcome the difficulty of time limits. The child that was discussed on the Waiting For Me program was abducted over 10 years ago and since then his face went through great changes. The only clue to find him was his baby pictures.  

Tencent created the cross-age facial recognition technology by utilizing its mass data to search for the abducted baby. It came up with the regularization transfer learning strategy based on distributed distillation learning (DDL) principles. The algorithm model is able to conduct the learning of cross-age facial recognition.  

 Dr. Li from Tencent Youtu Lab explains the facial recognition technology on the Waiting For Me program. [Photo provided to chinadaily.com.cn] 

Since 2018, Tencent Youtu has cooperated with Sichuan police and eventually set up a deep neural network model which has hundreds of layers of complicated structures following hundreds of model training and five version updates. The accuracy of the cross-age facial recognition has been as high as 96 percent.  

After comparing an abducted baby's pictures with the ones in its massive database, Tencent Youtu would select five children that most likely would be the missing person, and then DNA tests are conducted to decide on the right one, which saves much time and manpower for police during investigations.