While flexible bandwidth elastic optical networking is a promising direction for future networks, the spectral fragmentation problem in such a network inevitably raises the blocking probability and significantly degrades network performance. This paper addresses the spectral defragmentation problem using an auxiliary graph based approach, which transforms the problem into a matter of finding the maximum independent set (MIS) in the constructed auxiliary graph. The enabling technologies and defragmentation-capable node architectures, together with heuristic defragmentation algorithms are proposed and evaluated. Simulation results show that the proposed min-cost defragmentation algorithms can significantly reduce the blocking probability of incoming requests in a spectrally fragmented flexible bandwidth optical network, while substantially minimizing the number of disrupted connections.