Eye Detection in Image of Face with Mask from Various Viewpoints by Haar-based Cascade Technique

Abstract

Existing research on facial mask-worn face detection has limitations when applied to non-frontal face images. To improve the performance, side-view faces must be analyzed. The eye detection process is to analyze side-view face images. Therefore, we proposed an eye detection method from face images by a Haar-based cascade technique. Input data were images of frontal faces, non-frontal faces, face with mask, and face without mask. The face angle varied from negative 90 to positive 90 degrees. The proposed method consisted of the following process: 1) Database preparation, 2) Identification of head area and Center of the Head Calculation method installation, 3) a Haar-based Cascade Object Detector for Eye detection in angled faces preparation, 4) Eye side identification conditioning and 5) Implement of eye detection. To test the performance of the proposed method, 250 face images were used to measure the accuracy of each viewpoint of test images. In eye detection, the accuracy reached 86% for -90°, 90% for -45°, 88% for 0°, 90% for 45°, and 78% for 90°. The performance of the proposed method was satisfactory.