next up previous contents
Next: Texture-Mapped Models Up: Obtaining Texture Coordinates Previous: Obtaining Texture Coordinates

Deformable Mesh

Our approach exploits physics-based vision techniques for interactively localizing and tracking extended features in images, such as intensity edges. We employ the active deformable contours (popularly known as ``snakes'') introduced by Kass, Witkin and Terzopoulos Kass87. Deformable contours have been widely used in reconstructing 2D or 3D shapes from digital image data [Carlbom, Terzopoulos and Harris1994]. We connect a set of deformable contours to form a 2D deformable mesh. To obtain the texture coordinates for the fish body, we use a deformable mesh of the same dimension as the control-point mesh covering half of the fish body (i.e. tex2html_wrap_inline3010 ). Similarly, to obtain the texture coordinates for the fins, we use deformable meshes of the same dimensions as the corresponding control-point meshes of the fins. A deformable mesh floats freely over an image and it can be pulled interactively into position using the mouse. Fig. gif(a) shows the initial deformable mesh for capturing the texture of the body of a ``strip emperor'' angelfish.

The user begins by interactively dragging the border of the initial, regular deformable mesh such that it comes close to the intensity edges that demarcate the fish from its background in the image. Where necessary, the user may constrain any of the border nodes to selected anchor points on the edges using the mouse. Then the user starts a dynamic simulation that enables the border to approach and conform to the profile of the fish (i.e. the intensity edges) under the influence of an image force field. The remaining contours in the mesh relax elastically to produce a continuous, nonuniform texture map coordinate that covers the imaged fish body (Fig. gif(b)). Once the mesh has relaxed, the texture map coordinates are taken as the contour crossing points. In this particular example, since there are no dorsal or ventral fins, the body deformable mesh in fact covers the whole fish image. Appendix gif describes in more detail the mathematical formulation of the deformable mesh and how it functions.

  figure1062


next up previous contents
Next: Texture-Mapped Models Up: Obtaining Texture Coordinates Previous: Obtaining Texture Coordinates
Xiaoyuan TuJanuary 1996