Secure sharing of geospatial information has attracted widespread attention in recent years. As a typical geospatial data, vector maps can be used in many fields of national economic and social development. However, there are many security threats to data sharing of vector digital map, such as tampering, forgery, infringement, data leakages. Due to the complex spatial relation nature, existing data protection solutions are either infeasible or ineffective in such scenarios, especially for the exchange, sharing, and publishing of map data. To address these real security problems, in this paper, we propose a novel vector map protection framework, which introduces the idea of dynamic game and local desensitization. Two basic models and a specific scheme, including an important information protection model, an optimal data desensitization model, and a game-based local desensitization scheme, are presented in this framework. Dynamic game strategies can be used to effectively measure the utility of players under different decisions. Furthermore, important or sensitive map elements in vector maps can be desensitized locally in a controlled and reversible manner , thereby achieving more efficient data protection and secure sharing. A comprehensive evaluation, including desensitization strategies, performance and time consumption, accuracy control and reversibility, is given to show the security and feasibility of the proposed framework.