The double quantization effect of JPEG (Joint Photographic Experts Group) provides important clues for detecting image tampering. When an original JPEG image undergoes localized tampering and is saved again in JPEG format, the Discrete Consine Transform (DCT) coefficients of untampered regions would undergo double JPEG compressing, while the DCT coefficients of tampered regions would only undergo a single compression. The Alternating Current (AC) coefficient distribution accords with a Laplace probability density distribution described with a suitable parameter. And on this basis, this paper proposed a new double compression probability model of JPEG image to describe the change of DCT coefficients after the double compression, and combined the Bayes criterion to express the eigenvalues of the image blocks which have undergone the single and double JPEG compression. A threshold was set for the eigenvalues. Then the tampered region was automatically detected and extracted by using the threshold to classify the eigenvalues. The experimental results show that the method can detect and locate the tamped area effectively and it outperforms in terms of the detection result compared with the blind detection algorithm of composite images by measuring inconsistencies of JPEG blocking artifact and image forgery detection algorithm based on quantization table especially when the second compression factor is smaller than the first one.