Mixed reality applications can provide users with enhanced interaction experiences by integrating virtual and real world objects in a mixed environment. Through the mixed reality interface, a more realistic and immersive control style is achieved compared to the traditional keyboard and mouse input devices. The interface proposed in this paper consists of a stereo camera, which tracks the user's hands and fingers robustly and accurately in the 3D space. To enable a physically realistic experience in the interaction, a physics engine is adopted for the simulating the physics of virtual object manipulation. The objects can be picked up and tossed with physical characteristics, such as gravity and collisions which occur in the real world. Detection and interaction in our system is fully computer-vision based, without any markers or additional sensors. We demonstrate this gesture-based interface using two mixed reality game implementations: finger fishing, in which a player can simulate fishing for virtual objects with his/her fingers as in a real environment, and Jenga, which is a simulation of the well-known tower building game. A user study is conducted and reported to demonstrate the accuracy, effectiveness and comfort of using this interactive interface.