Social image aesthetic measurement based on 3D reconstruction

To measure the composition of photo and improve the aesthetic sense, this paper proposes a model that employs some heuristic principles for photographing in three-dimensional space. Compared with existing evaluation models, our approach takes advantage of the crowd-sourced three-dimensional information about where the photo was taken. Specifically, we first cluster the images with geo-tags from Internet to generate clusters of the same scene. To restore the geometry information of scene, photos with the same geo-tag are then combined together to generate a 3D model. In order to improve the performance of aesthetic measurement model, we replace the general saliency map algorithm, which is used universally to detect the salient region of the photo, with a method based on the 3D model to analyze the distribution of salient region. First, we perform the spatial clustering and calculate the mean of the frequency that its occurrences on cameras for each cluster. Through this method, it's available to quantify the saliency of each part in the scene and distinguish the specific part of the photo by calculating projection of the 3D saliency model, which is main theme of the picture. Based on a set of composition guidelines, including Dynamic Balance, Rule of Right Thirds and Diagonal Dominance, the composition aesthetics of each photo is estimated.