If features are given or can be defined they can be used to define a distance measure between objects. This can be understood as the euclidean distance in a properly scaled feature space. For good features this will also result in good dissimilarities. However, as long a dissimilarities are based on features their performance will be determined by the quality of these. As features do not describe the full objects it is possible that two objects that are different have a zero distance based on the available features. This is an essential cause of class overlap in feature spaces. Dissimilarities offer the possibility to overcome this. If the dissimilarity measure is defined in such a way that objects have a zero distance to itself or to entirely identical copies of themselves (which thereby should belong to the same class) there is no class overlap. … Dissimilarity Measure google