Considering the difficulty and problems in image matching for digital surface model (DSM) generation from high resolution satellite images (HRSIs) due to geometric projection and radial distortion, this paper has proposed an improved hierarchical image matching method based on triangle parallax constraints. In the method, scale invariant feature transform(SIFT) and normalized cross correlation (NCC) algorithms are used as well as epipolar geometric constraint and parallax constraint. Meanwhile, improvement is made for matching window adaptive optimization to generate DSM(digital surface model) from HRSIs based on matched feature points, grid points, and feature lines. Experiments were conducted to generate DSMs using HRSIs including WorldView-1 of Chongming island in Shanghai and ZY-3 of Zhoushan islands in Zhejiang province, and the effectiveness of the proposed approach was validated by control points and an existing DSM, with pixel level height accuracy achieved.