Ultradense networks have been identified as a promising technology to accomplish objectives of the fifth-generation wireless networks. However, the severe mutual interference generated by the densely deployed femtocells constitutes a great challenge. Different from the most prior studies that center on femtocell access points (FAPs) and neglect the influence of users' location when allocating subchannels, a centralized user-centric merge-and-split rule based coalition formation game, which can be well supported in the framework of the cloud/centralized radio access network, is proposed. This user-centric game makes it possible to utilize user information (e.g., distance) in estimating interuser interference so that the interference mitigation can be more accurate and effective. Besides, a novel resource allocation algorithm based on graph theory is presented. It can eliminate intratier interference efficiently by allocating users who may severely interfere each other in the conflict-graph with orthogonal subchannels as far as possible in a distance-aware sequence, and allocating subchannels in a profit-calculating method if idle subchannels are unavailable. Furthermore, in order to overcome the limitation that “only one subchannel can be allocated to each user” in previous coalitional games, a supplementary allocation algorithm is put forward to allocate remainder subchannels such that the system spectral efficiency can be improved. Simulation results show that the proposed algorithms improve the aggregate throughput up to 51.04%, 62.70%, 157.46%, and 482.42% comparing with the coalition formation game transmitted in a time-division multiple access (TDMA) manner within each coalition with virtual multiple-input multiple-output (MIMO) case, the reuse 1 case, the FAP-centric coalition formation game with modified recursive core case, and the coalition formation game transmitted in a TDMA manner among users with virtual MIMO case, respectively.