PROJECT TITLE :

A Flexible Framework for Exploring, Evaluating, and Comparing SHA-2 Designs

ABSTRACT:

As huge digitized fine art collections emerge and deep learning algorithms perform well, new research opportunities in the intersection of artificial intelligence and art are opening up. In order to investigate the applicability of deep learning techniques in understanding art images beyond object recognition and classification, we use convolutional neural networks (CNN) to predict scores related to three subjective aspects of human perception: aesthetic evaluation of the image, sentiment evoked by the image, and memorability of the image. For each topic, we evaluate numerous different CNN models trained on various natural image datasets and select the best performing model based on qualitative results and comparisons with existing subjective artwork evaluations. We also utilize a number of decision tree-based machine learning models to evaluate the relative relevance of various image variables including content, composition, and color in computing image aesthetics, visual sentiment, and memorability scores. Our findings suggest that both content and image lighting influence aesthetics, with color vividness and harmony having a significant impact on sentiment prediction and object emphasis having a significant impact on memorability. In addition, we look at the distribution of expected aesthetic, sentiment, and memorability scores across different artistic styles, genres, artists, and centuries in the context of art history. The method presented here enables new ways of assessing fine art collections based on highly subjective aesthetic qualities while also bridging the gap between traditional formal analysis and computational fine art analysis.