3D fountain modeling from single image

Transferring the objects that people use in their daily life to digital media has become a frequently needed process nowadays. One category of such applications are the transfer of historical buildings to digital environment by modeling in three-dimensions. Among the historical buildings, places of worship were very important in many cultures. Since the past, many glorious places of worship have existed until today. Mosques and fountains are important architectural structures in the Muslim culture. 3D modeling of these structures has become a concern of Cultural Heritage applications. In general, 3D modelling of an object requires relative pose of the camera for the different views, relation between the image points, and the corresponding line of sight and image points. There are some difficulties when performing 3D modeling even with images taken from different angles. With the advances in image processing and artificial intelligence, we focused on research on the ways in which an object can be modeled in 3D from a single image. Since the physical properties of each object are different from each other, it is necessary to produce a system specific to the object / structure to be modeled. In this thesis, a system is proposed to automatically generate 3D models of fountains from a single photograph. Although a single image is usually not sufficient for determining the 3D structure in an image; here we utilize our prior knowledge of general characteristic structures of fountains over the history such as the similarity of their top parts and their symmetrical shape. Firstly, the object is detected and labelled on the image using Convolutional Neural Networks (CNNs). Then, the object is segmented from background with Graph Cut technique and the contour of the object is used to estimate the shape of 3D model. Finally, 3D model of the fountain is completed by applying the texture acquired from the input image. Keywords: Computer Graphics, 3D Modeling, Convolutional Neural Networks, Deep Learning, Graph Cut, Cultural Heritage, Fountain.