Testing the contrast source extended Born inversion method against real data: The TM case