In the feature extraction part of a recognizer, we should be able to obtain the less complex representation as possible, in order to perform better in terms of computation and space requirements.
Face representations
There are many ways to create a model that can represent a face.
The most intuitive one is to represent the face as a two-dimensional image, modelled by a matrix that contains the pixels of the image.
It’s also possible to concatenate the rows in the image to represent the image as a vector of pixels.
A good way to represent an image is by associating a vector element for each feature that is extracted from the raw image. In this case we’ll obtain a multi-dimensional space that’s called Feature Space.