The objective of this master thesis is the recognition of human characters in TV-style video sequences. In order to recognize the same person at different time instances in a video sequence, the outward appearance of the person has to be described and learned with an appropriate model. The diversity in which humans can appear makes the task of human detection and character recognition to a particularly challenging problem. TVstyle movies provide an uncontrolled, realistic working environment for human detection and character recognition. Possible applications range from surveillance (e.g., intrusion detection) and security applications (e.g., person identification) to image retrieval or semiautomatic image annotation (e.g., automatic labeling of faces in personal photo albums).