Supervisors: Dr. Hayko Riemenschneider, Prof. Dr. Luc Van Gool
CAPTCHA systems are used in computing to tell humans and computers apart. Furthermore the collected input data can be used to solve problems a computer can not solve on its own. An example for such a problem is the matching of images. Today's image matching algorithms can match a wide selection of images but they still have their limits. Because the human brain is capable of dynamic thinking based on context, humans can often match corresponding objects of images which matching algorithms could not match. In this work we created a CAPTCHA to match features of two images. We compared the input data of humans with a SIFT based algorithm and an algorithm based on a deep, multi-layer, convolutional architecture to distinguish between humans and computers. Further we used the human generated feature matches to match image pairs which the matching algorithms could not match.