This R&D project was initiated to check base technologies for people identification.
To make it a little fun we decided to change requirements and define similarity or a person and celebrities photos from the open database. The result may be checked in Demo.
The following steps were done:
– face detection, with utilization of OpenCV Haar cascades and Dlib
– alignment of facial landmarks, with the help of OpenFace and bare heuristics
– face transformer, based on OpenCV functionality
– feeding to Neural Net, in order to get descriptor of face features, which can be compared with descriptor of other services
As a result, the solution has the pipeline of converting unique face features into descriptor, which can be compared with other persons, in order to get person with the most similar face features.