Aesthetic quality assessment of headshots

An automated system that can provide feedback about aesthetic value or quality of headshot photos based on learned rules could be a very useful support in photo searching, sorting and editing. This is a challenging problem as it requires semantic understanding of photos, which is beyond the state-of-the-art in computer vision. In this paper, we present a method built on most important rules or guidelines used by professional photographers to assess aesthetic quality of headshots. Proposed method uses low-level features and face-related high-level features. We make use of popular machine learning algorithms, support vector machines and Real AdaBoost, to determine whether a headshot is aesthetically appealing or unappealing. The results of extensive experiments indicate that proposed method is valid and effective: the overall classification accuracy for binary classification is greater than 86 %. This work is difficult to compare with previous attempts to assess aesthetic quality as no other research group studied this particular field of photography before.