Konferenční příspěvek

Kubečka L., Jan J.

Proceedings of the 3rd EMBEC Conference EMBEC'05

EAMBS

2005

256 - 261

17271983

Ing. Libor Kubečka (FEKT VUT Brno)

Kubecka_EMBEC [file]

Kubecka_EMBEC_Presentation [file]

The contribution deals with the processing of the 2D multi-spectral images,especially their fusion, using an orthogonal discrete wavelet representation of previously registered image pairs. The method is based on decomposition of gradient of the vector image by computation of eigen-values of the first differential of the vector image. An application of the method to multimodal ophthalmic images is presented. Two multimodal images (color fundus photograph and Heilderberg retina tomography image) are preprocessed and registered by the originally designed registration method based on maximization of the modified mutual information as the global similarity criterion. Controlled random search is used as the robust optimizer. 98% of images in the testing set containing 336 images have been successfully registered according to the subjective evaluation of the human observer. Further, results of the image fusion are presented as well as the first evaluation of segmentation of the optic nerve head contour.