Flaw detection plays a crucial role in many industries to make sure that the products meet the specified quality requirements. When making for example a car it is important that all the parts satisfy certain quality standards to make sure the consumer buys a car that is safe to operate. A crack or another weakness in a crucial part can be catastrophic. To make sure their cars are as safe as possible, car manufacturers are conducting thorough testing of crucial parts. Similar tests are done in a wide variety of industries, and these quality controls are often referred to as flaw detection. Any cracks, voids, or other weaknesses that can cause danger are called flaws. Flaw detection is often done, or preferred done, in real time-- in an assembly line fashion. An important constraint, in addition to reliability, is therefore speed. The techniques used in these tests varies. Common techn~ques are ultrasonic waves (1-D or 2-D), eddy current imaging, x-ray imaging, thermal imaging, and fluorescent penetrent imaging. In this thesis I will discuss automatic general purpose image-based flaw detection. "Automatic" means that the flaw detection is performed without human supervision, and "general purpose" means that the inspection is not tailored to a specific task (i.e. one particular flaw in one particular type of object), but is ideally applicable to any detection problem.