We propose a new method that obtains an inverse light transport matrix suitable for radiometric compensation; it creates seamless projection-based displays on unknown 3D structures. We extend the theory of inverse light transport to support uncalibrated projector-camera systems in which the optical axes of the projector and the camera are not aligned. The inverse light transport matrix consists of the inverse matrices of direct light transport matrix and inter-reflection matrix. Given that direct and inter-reflection matrices are separated from the light transport matrix, the inverse inter-reflection matrix is computed by Neumann inverse series. The proposed method introduces compressed sensing to optimally estimate the inverse direct light transport matrix from the sets of projector illumination patterns and the camera responses from which the global components are canceled by the inverse inter-reflection matrix. We visualize the matrix elements of the various light transport matrices generated by our approach to examine its performance. We demonstrate compensation trials that compare the proposed method with existing methods. Our experiments confirm that our method offers excellent projection-based displays on 3D objects.