Increasing the Training Set in Face Recognition Tasks by Using a 3D Model of a Face

Abstract

Abstract-The paper proposes a method for face recognition, which uses a 3D model of the head, which is turned at different angles to extend the size of the training set. Control points are placed on the images, which have been created using the 3D model of the head, and the distance ratios between the control points are stored in the database. Face recognition algorithm would look for two images of faces in the data base which have minimal difference between the turn angles of the head and the distance ratios.

Semi-automatic Face Image Finding Method, Which Uses the 3D Model of the Head for Recognising an Unknown Face

“…The method includes one existing algorithm for creating a 3D model of a head, which is described in [13], and is modified in [15]. This algorithm is used to create different face models from a base training set (profile, half turn and full face images).…”

“…Another method that was proposed to expand the training set [14], [15] included 4 new images of faces for each class. This approach also had its disadvantages: it could not precisely recognise the face at angles that were greater than 45 degrees.…”

“…The method includes one existing 3D model creation algorithm [13], [15] and the proposed algorithms that are based on distance calculation between control points, and searching for minimum square sums in order to recognise new face images.…”