IntroductionAs digital technologies advance, more and more publications are produced in digital formats and transmitted via the Internet. Accompanying such advance, however, unauthorized use, illegal copying, and malicious modification of digital products have become serious problems. Researchers thus try to find various ways to protect digital products; solutions include copyright assertion, content authentication, etc. In the area of image content authentication, the integrity of an image is regarded very important and must therefore be realized. A common approach is the use of digital watermarking techniques. Digital watermarking serves many purposes, for example, proof of ownership, content authentication, copy control, and so on.Researchers have developed various image authentication techniques to detect if an image has experienced unauthorized modification. Some of them can only detect whether the image as a whole has been altered. Others may have the additional capability to detect if a certain part of the image has been tampered with. Liu et al.[1] studied the Zenike moment values which are generated from low DWT subbands. They found that the quantized values are robust to common processing operations but fragile to malicious attacks. Therefore, they embedded the watermark by quantizing the Zernike moment values, and the locations (i.e., blocks) suffered from malicious attacks can be identified through examining the extracted values. Their method has moderate robustness against JPEG compression. In Rawat and Raman's scheme [2], two chaotic maps are used in order to enhance the security of the watermarked images. The pixels in the image are disturbed using the first chaotic map and are further separated into bit planes with the least significant bit used for watermark embedding. A binary watermark is scrambled by the second chaotic map. The watermarked images can avoid counterfeiting attacks. Xi'an [3] scrambled a bi-level watermark by the Arnold transform, and the Human Visual System is used to determine the quantization step. The scrambled watermark is then inserted into the low DWT coefficients. Tamper areas can then be localized by comparing the extracted and the original watermarks. Patra et al. [4] convert the images into the DCT domain and quantize the low-frequency coefficients according to the target levels determined by the Chinese Remainder Theorem. Their method is computationally efficient and is able to withstand such attacks as JPEG compression, sharpening, and brightening. Qi et al.[5] used two content-based watermarks to protect the images. One of them is generated by an edge detector for the purpose of detecting tiny changes, and the other is generated from the relationship between the wavelet coefficients for localizing tampered regions. Both watermarks are embedded into middle-and high-frequency DWT coefficients. Finally, the generated watermarks and extracted watermarks are compared to authenticate the image, and a malicious