Abstract: To improve the discrimination and invariance of feature descriptor, a robust overlapped gauge feature descriptor (ROGFD) was proposed. Firstly, the image grads was computed with Scharr algorithm and the second partial gauge derivatives were applied to enhance invariance of feature descriptor, and avoid the affection of Gaussian filtering on the feature descriptor efficient. Then the overlapped grids were constructed to enhance robust of feature descriptor by improving continuity of grids. Finally, the descriptor vector of length 64 was created by weighted summation and normalization of responses in neighborhood of key point. The experimental results of the recall vs. (1-precision) graphs and time show that, comparing with the state-of-the-art methods, the ROGFD reveals stronger robustness.