The latest publication by the CogNovo fellow, Francois Lemarchand, just appeared in Pattern Recognition Letters. The paper, named “Fundamental Visual Features for Aesthetic Classification of Photographs Across Datasets”, introduces a new approach to computational aesthetics.
The paper introduces a set of brain-inspired features existing in the human early visual system which can be extracted from images to determine people’s preferences. Francois developed algorithms to extract the distribution of orientations, curvatures, colours and symmetry, and couple them with one of the latest machine learning techniques, deep learning, to challenge performances of existing computationally-greedy solutions. Previous works learnt to recognise good-quality photographs by developing handmade scoring systems based on artistic rules, or using machine learning algorithms heavily to figure out patterns, but losing the possibility of gaining insight through feedback. The simplicity of the proposed system questions how useful high-level information, such as context or semantics, is to aesthetic judgement.
The research presented in this paper is the core building block of Francois’ thesis titled “Computational Modelling of Human Aesthetic Preferences in the Visual Domain: a Brain-Inspired Approach” and one of the results of his CogNovo PhD project.