5. Image detection

Here, we will only consider cases where a random process representation of the image proves effective for the functions we wish to implement. This field is still wide: anomaly detection, classification, etc., and we will limit the presentation to the question of anomaly detection, taking into account the specificities of the image:

• the "2D" nature of the image, which makes it impossible to find an order for the pixels that respects the notions of neighborhood (of the same dimension);

• the Poissonian character of the photon flux ;

• performance is limited by the presence of a background that comes from the scene and not from the camera's reception (BLIP).

Let's clarify these three points.