Dynamic object-based tomographic reconstruction

In this paper, we propose an object-based variational method for dynamic tomography. We model the dynamics of object boundaries with an affined transform and the dynamics of intensities with an autoregressive model. Based on these dynamic models, an energy function is defined which incorporates both the observed data, the prior knowledge about shape and intensity dynamics and prior knowledge about shape smoothness. A coordinate descent algorithm based on curve evolution methods is then developed to jointly estimate the underlying object boundary and intensity sequence. Efficient level set methods are used to implement the curve evolution process. In practice, we may not know the model of shape dynamics. We use our models to develop a gradient descent algorithm for the estimation of the dynamic model of shape from a training sequence of dynamic shapes. Preliminary experimental results demonstrate the strength of the algorithm in reconstructing dynamic objects from very sparse and noisy data.