A Kalman filter technique applied to surface reconstruction and visualization from noisy volume data

In this paper we propose a low-cost computational method applied to the smoothing of surfaces reconstructed from noisy data, as it is typically the case in ultrasound imaging. It makes use of tracking ideas. Our method is, to our knowledge, a novel and competitive alternative to those which make use of traditional methods of optical flow for the smoothing of the normals of an object's surfaces. Those methods, as it is well-known, are very involved in calculations. Our method is based on a Kalman filter; we propose a stochastic dynamic model which exploits the spatial coherence present in the data. We end up having a more efficient computational scheme with performance close to the optical flow method. A provision is made to impede the filter to diverge when the data depart from the assumed model. Our results both with synthetic and real Volume data show that our proposal is realistic in terms of rendering: a good trade off between computational resources and graphical results seem to be achieved.