Recently, numerous efforts were focused on 3D face models due to its geometrical information and its reliability against pose estimation and identification problems. The major objective of this work is to reduce the massive amount of information contained the entire 3D face image into a distinctive and informative subset interested regions based 3D face analysis systems. The interested regions are represented by nose and eyes regions of frontal and profile 3D images. These regions are detected based on distance to local plan descriptor only which is copes well with profile views of 3D images. The statistical distribution of distance to local plane descriptor is predicted using Gaussian distribution. The framework of the proposed approach involves two modes: training mode and testing mode. In the training mode, a learning process for local shape descriptor related to the interested regions is carried out. The interested regions (nose and eyes) are extracted automatically in the testing mode. The performance evaluation of the proposed approach has been conducted using 3D images taken from GAVADB 3D face database which consists of both frontal and profile views. The proposed approach achieved high detection rate of interested regions for both frontal and profile views.