A MEASURE OF 2D SHAPE-OF-OBJECT DISSIMILARITY

We describe an approach for measuring 2D shape-of-object dissimilarity . The shape of the different objects to be compared is mapped to a rep resentation invariant under translation, rotation, and area. Thus, the shapes will have the same amount of information to describe them (equ al number of pixels). The measure of dissimilarity is based on the tra nsformation of one shape into another. This transformation is performe d by moving pixels. Thus, the shape difference could be ascertained by counting how many pixels we have to move and how far to change one sh ape into another. When the shape transformation is performed, the dist ribution of the shape difference is computed, which permits an improve ment in shape comparison.