Relevant image pre-filtering in automatic image orientation correction

In this paper, we propose a method for improving automatic image orientation correction by relevant image pre-filtering. We find out that users feel more stress on false positives than false negatives in automatic correction through a user study. A strategy based on this finding is devised and our method realizes this strategy by filtering out irrelevant images which are likely to become false positives. The accuracy evaluations with actual personal photos show improvements of the precision from 94% to 98% at the recall ratio of 15%, and from 97.5% to 100% at the recall ratio of 10%. A subjective survey concludes the proposed pre-filtering alleviates the users´ annoyance for automatic orientation corrections.