A surface representation approach for novelty detection

There has been a pronounced increase in novelty detection research in recent years due to the driving force from applications such as monitoring of safety-critical systems and detection of novel objects in image sequences. This paper presents a novelty detection method from a new perspective by analysing the fundamental properties of novelty detectors. It constructs closed decision surface around the given data from known classes through the derivation of surface normal vectors and the identification of extreme patterns. A novel pattern is detected if it locates outside the region formed by the closed data surface. The experimental results demonstrate that the proposed method performs with high accuracies in detecting novel class as well as identifying known classes.