Method of Correction of Rotated Images Using Deep Learning Networks
Shun Ogiue, Hiroshi Ito · 2018
We propose a method of automatically correcting images by using deep learning networks regardless of metadata. If persons and scenes reflected in the eyes are inclined, human beings can recognize their inclination. This is thought to be because humans remember the inclination of images together with various events that occurred at the same time. Concurrent natural events are memorized in the human brain. Deep learning is a machine learning method that imitates the memory of concurrent natural events. The inclination of images is also one of the concurrent events. We propose a method of detecting and correcting the inclination of images using a classifier that has learned concurrent events by deep leaning algorithms.