Methodology for studying the influence of face rotation angle on the face detection accuracy

Abstract

The analysis of face detection methods in images was performed. For comparative analysis of efficiency, the Viola-Jones method based on the use of Haar cascades, the method of histograms of directional gradients and the method based on convolutional neural networks were selected. The angles of head tilt in the image were selected as variable parameters. The accuracy and speed of detector operation were determined as measured parameters. For the purposes of the work, the “Labeled Faces in the Wild” (LFW) face database and the "Head Pose Image Database" dataset were selected. Convolutional neural networks showed the best results.