The problem of novelty detection is considered for at set of dermatological image data. Different points of view are analyzed in detail. First, novelty detection is treated as a contextual classification problem. Different learning phases can be detected when the sample size is increased. The detection of the emergence a new class is considered here. A model that estimates the minimal amount of information required to recognize structure in the data as a function of class separability is also proposed. Secondly, texture alteration detection is considered a novelty detection problem. The possibility of avoiding pattern registration by transforming the data to a space invariant of registration is explored through a canonical analysis tool. Problems of expressing the data in a way they can be compared are also considered here, for instance, pattern registration. An approach for patterns alignment using an algorithm that iteratively approximateds the minimal value of an assumed smooth function is proposed. Registration outputs are evaluated and analyzed with the Multivariate Alteration Detection Transform.