Model based object recognition - The role of affine invariants

This paper proposes an efficient method to recognize rigid flat objects from its intensity images which are assumed to be arbitrarily positioned in space. The task of the recognition method is to find instances of known object models in affine images. Affine invariant shape descriptors of rigid flat objects are generated which are invariant to change in the point of view. In the proposed paradigm, the objects are described by sets of local and global features. Since we are also concerned with the recognition of partially occluded objects, the local features are given importance for obtaining descriptions of objects. The global features are useful for finding the exact match and are used for verification. The local features can be points, line segments, curve segments, etc. We restrict ourselves to points, which are referred to as interest points. The point set of the various model objects are matched simultaneously against the point set of the composite overlapping scene using a small number of corresponding points. Seven discrete moments are used here as global features which are also invariant under the affine transformation. Experiments show good performance and together with inherent parallelism of the recognition method makes the method a promising one.