------------------------------------------------------------------------------- B O S T O N U N I V E R S I T Y Computer Science Department C O L L O Q U I U M Thursday 14, 1994 11:00am (Coffee served at 10:30pm) Seminar Room / MCS 135 ------------------------------------------------------------------------------- Modal Models for Deformable Shapes Stan Sclaroff MIT I will describe ``modal models,'' a family of models that we have developed for representing deformable shapes. The modal representation is based on the finite element method; as a consequence, optimal estimates of object motion and shape can be made, and physical predictions and simulations can be computed directly from recovered models. Modeling and recognition performance is demonstrated on 3-D and 2-D data, found in grayscale images, X-rays, contours, optical flow, and range data. While the primary focus of this research is understanding 2-D or 3-D shapes in images, the formulation has useful applications in other areas: the sensor fusion problems found in aligning medical data; improved techniques for image metamorphosis; physical modeling for computer animation; and new methods for data compression. I will also describe results of my PhD thesis research in ``modal matching,'' a new method for matching and describing deformed shapes. In contrast to previous methods, we are able to compute our models directly from available image information, rather than requiring the computation of correspondence with an initial or prototype shape. This results in greater generality and accuracy, and is applicable to data of any dimensionality. Finally, I will demonstrate the utility of this approach for comparing shapes in images and for searching image databases. Host: Prof. Steve Homer ------------------------------------------------------------------------------- For more information contact Prof. Azer Bestavros -------------------------------------------------------------------------------