Master Assignment
AUTO: Auto encoders in face recognition
Type: Master EE/CS
Period: TBD
Student: (Unassigned)
If you are interested please contact :
Description:
Auto encoders (AE) are computational structures that have been used for image compression, restoration and more recently for image generation (Variational AE, Adversarial AE). Every type of AE contains an encoder that maps the input to a lower dimensional latent space and a decoder that maps the latent representation as close as possible back to the input of the encoder. Modern version of AE contain more components.
In some AE versions, the encoder output to the latent space can modelled as a draw from a conditional probability output. This enables to “guarantee” the probability distribution of same source and difference source pairs. A fundamental result in probability theory and statistics is the Neyman Pearson lemma: given the same and different source distribution, the likelihood ratio is the optimal classifier. Hence, in principle, an autoencoder can be used to create an optimal classifier.
Assignment
- Explore different network topologies for the AE.
- Explore the effect of using pre-determined facial features as an input (that is, calculated by OpenFace/ArcFace etc.), instead of facial images.