Image indexing and retrieval using image-derived, geometrically and illumination invariant features

We propose novel image-derived features for image indexing and retrieval in digital library applications. The new features capture the intrinsic geometry and color properties of an imaged object. That is, these features are insensitive to the change of an object's appearance due to incidental environmental factors such as rigid motion, affine shape deformation, changes of parameterization and scene illumination, and view point change. Recording both intrinsic shape and color information of an imaged object, these features allow the similarity between two objects to be accurately measured. Hence, they can be used to index objects of the same class with similar appearance (e.g., different airplanes) instead of objects of different classes (e.g., airplanes vs. automobiles).