For the application of state-of-the-art medical image processing methods to reality problems and their improvement, classical image processing still plays an important role, even though model- based approaches become increasingly important. In addition, medical image processing often integrates methods from machine learning. Many problems in medical imaging can only be solved by employing additional knowledge or priors – similar to a physician using his or her medical knowledge to interpret the data. This knowledge is generated from annotated images. We develop statistical models which incorporate the shape, the appearance and the physical properties of organs as well as the relative positions of different organs to each other. In addition to the pre-dominant CT, MR and ultrasound image data, also conventional 2D images from cameras play an important role. We develop computer vision based solutions for microscopy and dentistry and facilitate the matching of 2D images with radiological image data.