Video cameras have recently become common diagnostic tools in Magnetic Confinement Nuclear Fusion. They provide essential information for both the control of the experiments and the physical interpretation of the results. Since these cameras can produce up to hundreds of kiloframes per second and their information content can be very different, depending on the experimental conditions, several new image processing tools had to be devised to fully exploit these diagnostics. New structural pattern recognition algorithms have been developed to retrieve the required information from the large reservoirs of video frames in an efficient and reliable way. Specific real time algorithms, based on the computational paradigm of Cellular Nonlinear Networks, have been implemented on FPGAs to identify hot spots on the vacuum vessel and therefore to protect JET plasma facing components. Various machine learning tools, in particular Support Vector Machines, have been given Hu moments as input to automatically identify plasma instabilities. The methodology of the optical flow has allowed deriving information about the movement of objects in 3 dimensional space even if they have been detected by a single camera. A new anomaly detector based on an original interpretation of external support vectors is being tested with very positive results. Many of the more innovative solutions are based on quite general methods and are therefore expected to be applicable also in other fields of research.