Assessing perceptual data in images : a computational aesthetics approach

• Ano de defesa: 2021
• Autor(a) principal: Dalmoro, Bruna Martini
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul; Escola Politécnica; Brasil; PUCRS; Programa de Pós-Graduação em Ciência da Computação
• Palavras-chave em Português: Computational Aesthetics; Perception; Visual Features; Computação Estética; Percepção; Features Visuais; CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO
• Link de acesso: https://tede2.pucrs.br/tede2/handle/tede/10885

Human perception is the process that captures measurable physical stimuli and converts them into understanding information about the world. The study of human perception, regarding visual stimuli, is a wide area of research that has been studied in a multidisciplinar way. With the various advances in computing and the capacity for processing and analyzing images, human perception began to be studied computationally. One of the áreas that addresses this discussion is the area of computational aesthetics, a subfield of computational vision, which aims to research computational methods that can provide aesthetic decisions similar to those of humans. One of the applications of computational aesthetics today is the prediction of image and video ratings and their popularity. Another area much explored by computational aesthetics is the area of art and painting analysis. To build these algorithms, visual features drawn from images are used as a way to describe their content. The interpretability of these features is of great value for areas such as empirical and experimental aesthetics, as well as for generating insights from the found results. In the presente work, we explore three different problems contextualized in the computational aesthetic areas. In the first problem, we developed a model to predict the popularity of videos posted on Facebook using a dataset of visual features. In the second problem, we also use visual features and image category information (animation or live-action) to create a content-based movie recommendation system. In the third problem, we propose a methodology to identify and suggest influencing relationships between painters based on visual features extracted from the faces of their artworks. Our main objectives in this work are: to explore diferente problems involving different types of images from the point of view of computational aesthetics; to use only visual features to solve problems as a way to test the power and usefulness of this information in different applications; and to use only interpretable visual features to generate insights into the area of aesthetics and related areas. The results found in this work suggest that visual features, extracted from images and videos, are important resources for solving the proposed problems. In addition, the results indicate that the proposed methodologies are promising in trying to answer mathematically, in accordance with human perception, as intended by the area of computational aesthetics, questions about perception analysis. In addition, our methods allow to generate insights for aesthetic research when visual features are interpretable