Comparing Deterministic and AI-Based Approaches for 3D Reconstruction of Painted Scenes and Subjects

This paper discusses the development of two methods for reconstructing 3D models of painted or photographed scenes and subjects, particularly for the benefit of visually impaired individuals. In detail, two methods are discussed and applied to a set of case studies. The first method utilizes deterministic approaches paired with spatial scene reconstruction, to produce a tactile 2.5D model, or bas-relief representation, of the originally 2D scene. The second method implements artificial intelligence (AI) to create a 3D mesh from a 2D input of a painted subject. Both methods were tested against a set of case studies, including the reconstruction of subjects from paintings and photo printings. Results shows that the deterministic method is effective in reconstructing both subjects from paintings and even the entire depicted scene. However, the retrieved model is not fully 3D. Conversely, the AI-based approach produces rather accurate fully 3D models in a reasonable amount of time, but the models that are retrieved lack detail in a number of features, including the subjects’ face and clothing.