Most of the current stereo matching algorithms have high matching accuracy, but there are very few of them can realize real-time matching with video level frame rate. We present the multi-scale optimization algorithm based on Bayesian reasoning, which can be used to improve the matching accuracy, while maintaining the real-time performance. This algorithm obtains disparity maps with different scales by setting different window sizes. Based on this, the joint optimization based on Bayesian reasoning is proposed to optimize the disparity maps with scale information and complementarity. And then the high precision disparity maps are obtained. Test results of Middlebury stereo vision datasets show that the proposed algorithm has better accuracy and higher efficiency than several real-time algorithms.