Facial Matching

Facial matching is completed based on the two sets of critical facial features that represents personality traits respectively. An excellent score will be given to the unknown face during 1 : 1 or 1 : N facial authentication, if the registered faces had relatively similar facial features with the unknown face.

Age Estimation

Facial features are represented by the 68 unique points on a face. These facial features are able to accurately estimate the age of a person in a single capture. The technique is adaptable to changes of face angle in that capture, occlusion that may result from wearing mask or person's hands block part of their face, etc.

Gender Estimation

Gender classification has become relevant to an increasing amount of applications, particularly since the rise of social platforms and social media. With the same approach as in age estimation, the position of eyes, nose, mouth, etc, has been used to determine the gender of a person 

Anti Spoofing

Anti-spoofing technique with passive liveness detection which is non-intrusive. Users do not have to pay attention at all when they are being tested. The test is completed by several ways: single capture, sequence photography, thermal infrared image, etc.

Cross Platform

Core technologies are provided for free to let developer access the computer vision and image processing related SDKs that covers face detection, face tracking, face matching, anti-spoof, age gender multi-attributes and so on. It can also run different OS platforms like Linux, Windows, Android and iOS.

OCR Feature

High recognition rate with aid of modern deep learning technique and sophisticated rules. Support the extraction of personal information on physical card such as identity card, driving license card, credit card, etc; regardless of the character orientation either written horizontally or vertically. Support number plate recognition as well.

Object Recognition

Image classification model is built by supervised learning process with more than 8 millions sample images from nearly 1000 common objects. Users are allowed to perform transfer learning to optimise the model to recognise specialised object types, produce recognition rates higher than 90%.