Abstract: In order to make the outlier detection more automatic and decrease the users' difficulty for the selection of parameters, an outlier detection method with a new definition of average density is proposed. In this method, the outlier's density is considered smaller than the average density of data set and the none-outlier's density shouldn't decrease with its closed interval compression. An experiment is used to identify the outline of the animal's body. The experimental results show that the method identifies the face's outline effectively.