SMART FACIAL FEATURE REGIONS AND FACIAL FEATURE POINTS

Abstract

It is required to detect facial points automatically in order to make sentiment analysis, age analysis, gender identification, obtain three-dimensional virtual face model through two or more face images, develop augmented reality applications such as virtual makeup. In the present study, an integrated algorithm and a software were developed to find position of human faces, facial feature regions and facial feature points using standardized images. Images can be taken from different angles and poisitions. The data used for testing purposes is obtained from a facial database that contains human face images with 1600x1200 resolution taken by Inspec Mega Capturor II 3D (optical 3D digitizer) structured - light 3D digitizer device. Some cases were excluded from the scope of this work such as glasses, beard and mustache, different skin colors, different emotions, facial expressions. These objects cover some parts of face and require preprocess operations. In the present study, color analysis methods and Haar classification methods are integrated. In testing stage 35 different human face images contain 11 men and 24 women, with resolution 1600x1200 pixels. 100% capture face accuracy is obtained as a result of the tests. Mean square error is calculated as 2.04086 pixels. Using smart feature regions and feature points, 3D face point cloud was produced and stereo face images are matched easily.