An interactive Platform for Representing, Interlinking and Analyzing Content on Art

• Autor(a) principal: Ribeiro, Guilherme Figueiredo
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/151777
• topic: Art; Platform; Machine Learning; Algorithm; Catalog; Interlinking; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Unfortunately, many art objects such as paintings and sculptures are not yet cataloged in digital form. Often not even in printed catalogs or books. This circumstance provides a considerable obstacle for obtaining information from different collections within mu- seums or private collections all over the world. Because a significant number of works of art are also never cataloged, either because the owners lack the knowledge to do so or because there isn’t a simple tool to help the cataloging process, by providing a large amount of information in an accurate way, a significant number of these works of art also remain unknown to the world. With the help of machine learning algorithms and tools that can interpret images contents, a platform is to be created that offers a simple data acquisition from art pieces. In addition to simpler technical attributes such as style and type of art, more detailed information such as probable artist, materials, and decade of creation could already be suggested to the user during the data acquisition. This is possible thanks to the algorithms that can analyze images of the object in question. Alongside these suggestions, there would be the possibility to interlink data by multiple features in order to get similar art pieces to the submitted one. From the user utilization of the platform, the growing data set shall be linked with each other, visualized in an interactive platform, and finally serve to improve the analysis of new art objects. The creation of this platform made it feasible to offer a wide amount of information about any work of art. Four separate image classification models that could categorize art pieces based on their author, style, decade, and type were developed. To improve the prediction power of these, a number of strategies were studied and applied. Further details, including a set of relevant tags, the most dominant colors, and the materials utilized in the pieces of art, were also provided using Google Cloud Vision API. The platform’s whole infrastructure was built with these features and best practices in mind.