A new use of the ridgelets transform for describing linear singularities in images

Ridgelets transform, related to the wavelets and the Radon transform, offers a sound mathematical framework to organize linear information at different scales of resolution. We use this multiresolution property to define a representation scheme at several levels of decomposition to describe images whose relevant information is concentrated around linear singularities. We have also defined a classification procedure that combines information from all levels and is invariant to usual affine transformations. We have applied this scheme to the description and classification of linear graphic symbols, achieving recognition rates over 97% with the GREC’03 symbol database, even with degraded and distorted images.