Algorithmic agents in sound, image and video (Michael Murtaugh & Nicolas Malevé, Constant)

Title of film with blurry actor????

you learn about an image through its transformation
* for contour detection: 
    start with blurring edges
    not coherent lines, guess which fragments belong together to create a shape
* face recognition Volterra Kernel Training
mass of documents available -> allows statistical model of a face
out of a lot of manual work done by humans, every detail of face (with different lightnings) is annotated --> working conditions/labour army
-> training data needs also images where the face is hidden / no image exists in isolation

False positive
images selected as containing a face, but don't
(side note: see interesting iphoto bug here: )
feedback loop: hiding for the algorithm
same model can be used for other shapes


(external but related topics)
on recognition algos