Milos Stojmenović
The seminar presentation is in the field of image processing. We are interested in real time object detection in images (jpegs). By real time, we mean the computer should be able to process roughly 8 images/second. There are several approaches to solving such image retrieval problems. We will discuss the machine learning based image retrieval approach by Viola. In his face recognition scheme, two libraries are given as input: a library of positive images (faces) and a library of negative images (non-faces). By applying integral images, Adaboost machine learning and cascading, a decision procedure is created. The tested image is then entered into a decision procedure to report the presence or absence of faces in the image. The Adaboost method will be described, and its applications will be mentioned (Viola face ). The main goal of the s thesis is to see how well this learning based approach will work on a different set of objects, such as cars or car license plates. The seminar presentation will be based on my Master's thesis.