Detecting orientation of in-plain rotated face images based on category classification by deep learning
Yoshihiro Shima, Yumi Nakashima, Michio Yasuda · 2017
Digital cameras and smartphones with orientation sensors enable portrait images to rotate automatically. This is done by using an image file's metadata, which is in the exchangeable image file format (EXIF). The output of these sensors is used to set the EXIF orientation. Unfortunately, software program support for this feature is not widespread or consistently applied. Our research goals are to create an EXIF orientation flag for detecting the upright direction of face images having no orientation flag and to apply software for organizing photos. In this paper, we propose a novel orientation detection method for face images that relies on image category classification by deep learning. Rotated images are classified in four classes, namely 0°, 90° clockwise, 90° counter-clockwise, or 180°. As an image feature extractor, a pre-trained convolutional neural network is used, and the support vector machine is used as a classifier. The conventional part-based face detection method that uses Haar-like features is compared with the proposed orientation detection method based on deep learning. Experimental results on 450 face image samples show that the proposed method is very effective in detecting the orientation of face images with background variations.