Qualitative Portrait Classification

Due to recent advances in high-quality digital pho- tography, taking a large series of images is very in- expensive. Especially in portrait situations, this re- sults in a possible advantage because subjects of- ten feel uncomfortable during acquisition. Select- ing from a larger set of images increases the chance of a more satisfying outcome. However, the selec- tion process is not easy and time consuming as only a small number of images is typically considered as aesthetically pleasing. In this work, we propose a machine learning approach to mimic the selection process of a human subject. After a short training period, a large set of images can be classified in- stantly into two categories, good or bad. With the proposed automatic pre-selection, the advantage of digital photography for portrait images is brought to a new level.