2.5D Modeling from Illustrations of Different Views

When artists design characters, they draw illustrations viewed from the front, side or slant views. In this paper, we create a 2.5D model using such illustrations. A 2.5D model is a model that is created by arranging deformable billboards along the depth direction, and it can express appearance of the character between two viewpoints with considering depth information. Our method uses two cartoon-like illustrations and the corresponding eye directions as inputs. These illustrations are composed of contours and closed regions painted with uniform colors. Given closed regions in illustrations, our method finds corresponding closed regions between two illustrations based on an improved similarity function, which yields better matching than a previous method. After obtaining correspondences of closed regions, our method creates a 2.5D model by transforming each pair of matched regions as a 3D billboard whose position can be estimated based on the correspondence. These processes are semi-automatic. The user then assigns feature points manually along the corresponding contours of matched regions so that the contours can be interpolated naturally. Finally, by completing regions that appear or vanish with user strokes, our method can interpolate illustrations with occluded regions. We demonstrate that our method can create 2.5D models with various illustrations.