Constructing the feature descriptor plays a vital role for the keypoint-based image registration method. The existing local descriptors primarily depend upon the image local texture information. Since high-resolution remote sensing images contain much repetitive texture, the local descriptor-based approaches usually suffer from false matching due to the similar keypoint descriptors. To tackle this problem, this paper proposes a novel keypoint matching method for high resolution remote sensing image registration based on the local and global keypoint descriptor. First, to enhance the robustness of the keypoint, we build a LG-SIFT (local and global scale invariant feature transform) descriptor relying upon the image local texture information and the keypoint global distribution, and then design a similarity measure given consideration to the local and global representation of the LG-SIFT descriptor. Next, the probabilistic relaxation labeling approach is employed to the keypoint matching, in which the compatibility coefficient is established by the keypoint similarity measure. The experiments carry out on various high resolution remote sensing images, and the experimental results demonstrate the effectiveness of our method while dealing with the images of low overlapping area, high repetitive texture, and big geometrical deformation.