APPLICATION OF A MULTI-LAYER PERCEPTRON NETWORK FOR DETECTION OF HUMAN FACES ON A LARGE DATABASE
Metin Salihmuhsin, Yavuz Selim İşler · Erzincan Üniversitesi Fen Bilimleri Enstitüsü Dergisi · 2015
In this paper, we have designed and developed a face detection system that could detect human faces of different sizes and orientations in gray scale image files.The developed system consists of a camera, a computer, an image acquisition setup and a face detection method written in Matlab.In order to test detection capability of the system, we have collected a database that contains 125 different images of 125 different people with the above mentioned image acquisition setup.A program based on a multilayer perceptron network (MLP) is developed in order to detect faces in images.The MLP based face detection program works in two levels and uses the symmetry inherited in a human face between left and right sides.In the first level, the detection system searches for a similar structure to a right side of a human faces in the image using a predefined face template in multi resolution.When the detection method finds such a region, it processes that portion of the image further in order to find whether the other side of the face exists or not in the closed vicinity.We have tested our system on a database that contains 100 images selected from above mentioned database.The remaining 25 images are used to form training set for the MLP algorithm.Simulation results show that the method performs %81 correct detection rate on the test set.