This paper proposes an algorithm to decode the image steganography using entropy features. When the original image of steganography is randomly permutated in the rows and columns, it is difficult to decode the stego image into the original and cover images without correct random permutation matrix. This paper exploits the entropy features in the decoded images, where the local entropy distributions of incorrectly permutated images are different from that of correctly permutated image. According to the proposed decision rule based on local entropy distribution, randomly permutated stego images can be correctly decoded without supervised decision.