The traditional blind detection methods of image copy-paste forgery are time consuming, of high computation cost and low detection precision. A blind detection algorithm of copy-paste image forgery based on Mean Shift (MS) was proposed in this paper, which extracted Speed Up Robust Feature (SURF) points and then performed feature matching utilizing the method of best bin first in order to filter redundant points and locate the copy-paste forgery regions preliminarily. Pixels with the same or similar attributes would be segmented in the same region after implementing MS. The copy-paste regions could be detected according to the position dependency between matched feature point with its segmented region of MS and the detection result would be further refined by comparing the similarity of edge histogram and HSV (Hue-Saturation-Value) color histogram among the segmented regions of matched SURF and its neighborhood, and those regions with large similarity were included in the forged region. The experimental results show that the copy-paste forgery regions are detected accurately in the image with clear outline and rich details, and the proposed algorithm can robustly and efficiently detect the copy-paste forgery regions of image.