How to define the quality of images

Juan Romero,Adrian Carballal,Nereida Rodriguez,Iria Santos,Luz Castro

doi:10.3390/mol2net-04-05504

Abstract

One of the main retoes in Artificial Intelligence is to do creative tasks such us creating music, art or novel ideas. But, even more "simple" task such us have aesthetic taste is very hard to a computer system. In this paper we will present some steps into the building of a system that learn to predict the aesthetic and quality value. We will also show preliminar results into a experiment where human evaluate different images based on its quality and aesthetic value. Finally we will show some of the posibilities of this tecnology.

Highlights

One of the main challenges in Artificial Intelligence is to do creative tasks such us creating music, art or novel ideas
This new dataset was evaluated by two different populations, creating several models in Machine Learning for the automated prediction of aesthetic and quality value, as well as that of DPChallenge.com
It has been evaluated in 3 different ways: evaluation from the DPChallenge portal with at least 100 scores per image; an aesthetic evaluation conducted under controlled experimental conditions and a minimum of 10 votes per images; and a quality evaluation conducted with similar conditions

Summary

Introduction

One of the main challenges in Artificial Intelligence is to do creative tasks such us creating music, art or novel ideas. Introduction (optional) Several research groups have tried to create computer systems capable of learning the way a group of human beings perceive aesthetics, to make a generative system or use for the selection or an automatic order of images. Another type of studies were those of Cela-Conde et al [5], Forsythe et al [6] and Nadal et al [7] with a small set of images, but evaluated under control by a group of people with experimental conditions.

Results

Conclusion