In many applications of tomography, such as electron microscopy, industrial non-destructive testing, cardiac imaging, etc, the aim is to find the components constituting the object. Traditional approaches would first reconstruct the density distribution from the projection data and then segment (label) this distribution. This book introduces a new paradigm of directly estimating the label image from the projections, by postulating a low level prior knowledge regarding the underlying distribution of label images. Because of the typically small number of labels, this problem offers significant challenges and opportunity: much fewer data is required as a result. This work provides strategies, algorithms, as well as methods for choosing suitable Gibbs prior. Anyone who may be considering reconstructing label images from limited data should find this book a useful guide.