# Survey of Face Recognition Techniques under occlusion

###### Tags: `Face Recognition`

## Problems with Face Recognition systems
Face recognition models struggle to perform in conditions of pose variation, vaiation in illumination, low resolution images, diffent experessions and occlusions.

#### Why is occlusion a problem?
We don't have prior knowledge of what kind of occlusion will be there in the test images. Also it isn't possible to collect occluded face dataset and train on it and expect the model to work fine for all kinds of occlusion.



#### Types of occlusions for testing OFR(occluded face recognition)

• Real occlusions: gallery images are mugshots free from
occlusion while probe images are faces occluded by
realistic images such as sunglasses, or a scarf.
• Partial faces: gallery images are mugshots free from
occlusion while test face images are partial faces; hence
the name partial face recognition is given by researchers.
• Synthetic occlusions: gallery images are faces in the wild
which are captured from uncontrolled scenarios while
probe faces are blocked with synthetic occlusions to
simulate real occlusions.
• Occluding rectangle: gallery images are occlusion-free
mugshots while test face images are occluded with a
rectangle such as white and black rectangles.
• Occluding unrelated images: gallery images are
mugshots free from occlusion while test face images are
occluded with unrelated images such as a baboon, or a
non-square image.


## Occluded Face Detection
> Here we are discussing detection and NOT recognition

Face detection with occlusion can be briefly classified into the following categories.
* General Face detection: These are methods that were developed to detect normal images(non occluded ones)
* Occluded Face Detection: