Multi-modal registration is the task of aligning images from an object acquired with different imaging systems, sensors or parameters. The current gold standard for medical images is the maximization of mutual information by computing the joint intensity distribution. However intensities are highly sensitive to various kinds of noise and denoising is a very challenging task often involving a-priori knowledge and parameter tuning. We propose to perform registration on a novel robust information support: the wavelet energy map, giving a measure of local energy for each pixel. This spatial feature is derived from local spectral components computed with a redundant wavelet transform. The multi-frequential aspect of our method is particularly adapted to robust registration of images showing tissues, complex textures and multiple interfaces. We show that the wavelet energy map approach outperforms the classical framework in rigid registration experiments on synthetic, simulated and real data, whether noise is present or not.     «

Multi-modal registration is the task of aligning images from an object acquired with different imaging systems, sensors or parameters. The current gold standard for medical images is the maximization of mutual information by computing the joint intensity distribution. However intensities are highly sensitive to various kinds of noise and denoising is a very challenging task often involving a-priori knowledge and parameter tuning. We propose to perform registration on a novel robust information s...     »