Image Aesthetics

In this paper we investigate the use of a deep Convolutional Neural Network (CNN) to predict image  aesthetics. To this end we finetune a canonical CNN architecture, originally trained to classify objects and scenes, by casting the image aesthetic prediction as a regression problem. We also investigate whether image aesthetic is a global or local attribute, and the role played by bottom-up and top-down salient regions to the prediction of the global image aesthetic. Experimental results on
the canonical Aesthetic Visual Analysis (AVA) dataset show the robustness of the solution proposed, which outperforms the best solution in the state of the art by almost 17% in terms of Mean Residual Sum of Squares Error (MRSSE).

Publications

1.

Predicting Image Aesthetics with Deep Learning
(Simone Bianco, Luigi Celona, Paolo Napoletano, Raimondo Schettini) In Advanced Concepts for Intelligent Vision Systems: 17th International Conference (ACIVS 2016), volume 10016, pp. 117-125, Springer International Publishing, 2016.

Download BibTex Doi 

@inproceedings{Bianco2016predicting-image,
 author = {Bianco, Simone and Celona, Luigi and Napoletano, Paolo and Schettini, Raimondo},
 year = {2016},
 pages = {117-125},
 title = {Predicting Image Aesthetics with Deep Learning},
 volume = {10016},
 publisher = {Springer International Publishing},
 isbn = {978-3-319-48680-2},
 booktitle = {Advanced Concepts for Intelligent Vision Systems: 17th International Conference (ACIVS 2016)},
 pdf = {/download/Bianco2016predicting-image.pdf},
 doi = {10.1007/978-3-319-48680-2_11}}