Semi-fragile watermarking which is robust to given operations and fragile to conscious operations, is a useful authentication technique for digital media. In this paper, a novel semi-fragile image watermarking algorithm that accepts gradient sharpening is proposed. In the proposed algorithm, based on the theory of gradient sharpening, the watermark is embedded in the space domain by modifying the parity of the pixel value and its gradient, so it has very low complexity and high fidelity. The robustness and frangibility of the proposed algorithm to given image processes are analyzed theoretically and experimentally, and the results show that the proposed algorithm is very robust to gradient sharpening with any threshold, while it is very fragile to other attacks, such as Gauss noise, cropping, neighborhood averaging and median filter. Namely, the proposed algorithm can distinguish gradient sharpening from other attacks accurately.