Automatic Single Person Composition Analysis

Abstract

This paper presents an approach to image analysis based on three photographic composition rules: Rule-of-Thirds, Zoom Rule and Integrity Rule. These rules are commonly used by experienced photographers as an important step for creating attractive photos. The proposed approach assumes there is only one person in the image and considers the use of a face detector to locate the photograph’s main subject. The composition analysis computes a set of numerical measures from an input image. These measures are then combined to produce an estimate for the overall composition quality. Experiments involving a subjective evaluation by human observers have demonstrated promising results, given that some correlation has been observed between labelling by expert users and the proposed automatic analysis in up to 85% of a test set of images.