In this article, a real-time object tracking method based on perceptual hashing called MHash is put forward. In order to enhance the robustness of noise and the apparent change, hash down-sampling is employed to ignore the image aspect ratio and gain low frequency information. Solving the problem of partial illumination change and occlusion, fragment-hash tracking is proposed. Overcoming the limitation of global illumination change, the mean-compared method is used. Also, integral image is applied to accelerate the calculation of the mean and confidence map to filter matching noise for the maximum inhibition. Experimental results show that MHash algorithm has very strong robustness especially for partial occlusion and illumination change, and achieves accurate tracking in a variety of sequences. The average tracking speed reaches 52.12 fps that shows high speed.