The vibration and friction factors of two-axis motion platform cause video jitter and because of image feature extraction and matching influenced by illumination variation, noise and so on, the wrong motion vector will be estimated. The multi-scale space two dimension high precision video stabilization algorithm is proposed based on concentration degree. Invariant features are extracted, and then the depth first nearest neighbor search algorithm is used to get the matching double points. At last, concentration degree is computed, and different computing models are built to estimate compensation parameters by optimal concentration degree principle. 320×240 pixels gray video sequence is used in experiment, and the algorithm proposed in this paper has high precision. The estimation time is just 13% of Random Sample Consensus (RANSAC), which improves the performance of video stabilization.