Relatively high radiation CT techniques are being widely used in diagnostic imaging raising the concerns about cancer risk especially for routine screening of asymptomatic populations. An important strategy for dose reduction is to reduce the number of projections, although doing so with high image quality is technically difficult. We developed an algorithm to reconstruct discrete (limited gray scale) images decomposed into individual tissue types from a small number of projections acquired over a limited view angle. The algorithm was tested using projection simulations from segmented CT scans of different cross sections including mid femur, distal femur and lower leg. It can provide high quality images from as low as 5-7 projections if the skin boundary of the cross section is used as prior information in the reconstruction process, and from 11-13 projections if the skin boundary is unknown.