Automated detection of cells in phase contrast images Tanja Schilling Tanja.Schilling@t-online.de Interview and presentation for PhD position Abstract: A method is presented that detects cells in phase contrast images of cervical smears which are used for early diagnosis of cervical cancer. The detection is done in two stages. At the first stage, image regions which contain epithelial cells are found using texture features. The exact contours of those cells are identified at the second stage by applying an active contour model. Statistical Geometrical Features were chosen as texture features. The most representative features were identified using Sequential Forward Floating Selection. A Kernel Fisher Discriminant is applied to this feature set to classify the image regions as cells or irrelevant background. The method was tested on a set of 69 images in which 95% of the image areas were identified correctly as cells. The presented method was content of a Diplom (i.e. MSc) thesis.