In this paper, a spread-spectrum-like discrete cosine transform (DCT) domain watermarking technique for copyright protection of still digital images is analyzed. The DCT is applied in blocks of 8x8 pixels, as in the JPEG algorithm. The watermark can encode information to track illegal misuses. For flexibility purposes, the original image is not necessary during the ownership verification process, so it must be modeled by noise. Two tests are involved in the ownership verification stage: watermark decoding, in which the message carried by the watermark is extracted, and watermark detection, which decides whether a given image contains a watermark generated with a certain key. We apply generalized Gaussian distributions to statistically model the DCT coefficients of the original image and show how the resulting detector structures lead to considerable improvements in performance with respect to the correlation receiver, which has been widely considered in the literature and makes use of the Gaussian noise assumption. As a result of our work, analytical expressions for performance measures, such as the probability of errors in watermark decoding and the probabilities of false alarms and of detection in watermark detection, are derived and contrasted with experimental results.

In this paper, we address the problem of the performance analysis of image watermarking systems that do not require the availability of the original image during ownership verification. We focus on a statistical approach to obtain models that can serve as a basis for the application of decision theory to the design of efficient detector structures. Special attention is paid to the possible nonexistence of a statistical description of the original image. Different modeling approaches are proposed for the cases when such a statistical characterization is known and when it is not. Watermarks may encode a message, and the performance of the watermarking system is evaluated using as a measure the probability of false alarm, the probability of detection when the presence of the watermark is tested, and the probability of error when the information that it carries is extracted. Finally, the modeling techniques studied are applied to the analysis of two watermarking schemes, one of them defined in the spatial domain, and the other in the direct cosine transform (DCT) domain. The theoretical results are contrasted with empirical data obtained through experimentation covering several cases of interest. We show how choosing an appropriate statistical model for the original image can lead to considerable improvements in performance.

In this paper we analyze the effect that the application of channel coding produces on the performance of the watermark detection and decoding tests for copyright protection of images. Detector structures are derived for both tests and analytical bounds and approximations are obtained for the bit error rate (BER) and the receiver operating characteristic (ROC) associated with the watermark decoding and detection tests when block codes are employed. The extension to other families of codes is discussed. Finally, the analytical expressions are contrasted with experimental results in several cases of interest.

A watermarking scheme for copyright protection of still images is modeled and analyzed. In this scheme a signal following a key-dependent two-dimensional multipulse modulation is added to the image for ownership enforcement purposes. The main contribution of this paper is the introduction of an analytical point of view to the estimation of performance measurements. Two topics are covered in the analysis: the ownership verification process, also called watermark detection test, and the data hiding process. In the first case, bounds and approximations to the receiver operating characteristic are derived. These results can be used to determine the threshold associated to a required probability of false alarm and the corresponding probability of detection. The data hiding process is modeled as a communications system and approximations for the bit error rate are derived. Finally, analytical expressions are contrasted with experimental results.