Image orientation estimation with convolutional networks

Abstract

Rectifying the orientation of scanned documents has been an important problem that was solved long ago. In this paper, we focus on the harder case of estimating and correcting the exact orientation of general images, for instance, of holiday snapshots. Especially when the horizon or other horizontal and vertical lines in the image are missing, it is hard to find features that yield the canonical orientation of the image. We demonstrate that a convolutional network can learn subtle features to predict the canonical orientation of images. In contrast to prior works that just distinguish between portrait and landscape orientation, the network regresses the exact orientation angle. The approach runs in realtime and, thus, can be applied also to live video streams.