Lecture: Fooling the Computer Eye
Manipulating images to make an algorithm misclassify them
In the field of image classification, algorithms are trained to properly label images. What if we actually wanted to fool an algorithm into misclassifying an image?
We will walk through methods to perturbate an image (single pixels, patterns, noise) in such ways that it is still recognizable for the human eyes, but is not properly classified by a trained classificationmodel anymore once it has been perturbated.