Local stereo matching methods still play an important part as they are simple and fast. Some local methods perform well and even better than most global methods. But they usually achieve accuracy at the expense of speed. Simple local methods are fast, but exhibit systematic errors. In this paper, we focus on the invalid regions of traditional window-based matching and present a new solution to improve the initial matching result according to the identification of invalid regions and weight-based disparity adjustment. Our solution provides an easy way to increase the accuracy with little additional computation. The experimental results are evaluated using the Middlebury stereo test bed, showing that our method achieves competitively performance. Moreover our algorithm modules could be easily expanded into other matching algorithms as post-processing.