Application of convolutional deep networks into face recognition systems reduces some of the problems with face alignment in images that have significant impact on the results in the previously used methods. However large variation in head pose still might be the issue even for modern algorithms. It is natural to think of applying 3D image processing methods to solve the problem of algorithms reliability over pose changes. Solutions based on native 3D data like point clouds and voxel grids can be infeasible because of heavy structures and more difficult to collect databases. One of the recently proposed approaches is to decompose the object into the set of multiview 2D images. For faces, collecting the sets of different view images is also typical for police booking photography. In this talk she will analyze some of the multiview recognition methods, like MVCNN and RotationNet, in application to face identification databases.
Curriculum vitæ
Weronika Gutfeter is a research assistant in Biometric and Machine Intelligence Laboratory of Research and Academic Computer Network (NASK) in Warsaw, Poland. She obtained her Master’s degree in Computer Engineering at Warsaw University of Technology and now she works on her PhD thesis that covers a topic of 3D, semi-3D and multiview face identification. Her research interests include also other computer vision methods like algorithms for human detection in depth and thermal images. In NASK she works on developing machine learning systems for government and public sector.
We thank our media partner:
