Abstract: A perceptual model performs an analysis of a media signal, such as an image or audio signal. The model may be used in media signal processing applications such as digital watermarking and data compression to reduce perceptibility of changes made to code the signal. For image applications, the model computes the sensitivity of an image to changes based upon local image contrast, while taking into account the sensitivity of connected directional edges. By comparing the local image strength of various directionally filtered versions of the image, the model creates a directional control vector. This control vector may be used to reduce changes to an image in text and edge regions, and thus, avoid perceptible artifacts in those regions. The model takes into account the local contrast of the image and the directional control vector to create a gain vector. Using the local contrast measurements, the model follows the eye's nonlinear response to contrast discrimination.

Abstract: A watermarking method converts a watermark message into a Frequency Shift Keying (FSK) signal and embeds the FSK signal in a host signal. The spectral properties of the FSK watermark signal facilitate its detection, even in applications where the watermarked signal is corrupted. Because of these properties, the FSK watermark signal can perform the dual function of identifying the watermark's presence and orientation in potentially corrupted media, and also conveying a hidden message in the host signal. Such a watermark may be referred to as a self-orienting watermark.

Abstract: The disclosure describes methods for adaptive prediction filtering for digital watermarking applications. The predictive filter is designed to predict the digital watermark signal. In one implementation, there are two stages of prediction filtering. A first stage operates on a suspect signal before detecting any watermark in that signal. This stage adapts the predictive filter to local signal characteristics. A second stage uses known watermark signal elements to adapt a non-linear filter used in extracting message symbols from the suspect signal. This stage produces detection statistics based on a comparison of expected and measured message symbol values, and uses these statistics to adapt the parameters of a predictive filter. This adapted prediction filter is then applied to extract message symbol estimates.

Abstract: A method for tracking media signals in multimedia content is provided by uniquely identifying the multimedia content with a digital watermark identifier. A video channel is preferably embedded with a first watermark identifier. The first identifier uniquely identifies the multimedia content or the content family. A second digital watermark is embedded in an audio channel of the multimedia content. The second digital watermark uniquely identifies a user or user device. When the multimedia content is found in an unauthorized or unexpected channel, the content can be traced back to its misappropriating source via the user identifier.

Abstract: A method of synchronizing a digital watermark detector divides a watermarked signal into blocks, each block including a portion of a watermark signal. For each block, the method computes a local correlation space comprising a neighborhood of correlation values by correlating the watermarked data in the block with a known watermark signal at a neighborhood around the block. It then finds a correlation maxima in the local correlation space for each block, where the correlation maxima indicates a local offset used to align the watermarked data in the block before decoding a watermark message from the block. To further refine the synchronization, the method filters the array of local offsets for the blocks to provide a refined set of offsets. The offsets and corresponding correlation values provide an indicator of where message symbol decoding will provide more accurate results, enabling the watermark message decoder to selectively decode symbols from portions of a signal block containing a watermark signal.

Abstract: A method of removing fixed pattern noise derives an estimate of fixed pattern noise from a composite media signal and uses the estimate to evaluate and remove fixed pattern noise from selected frames. The technique is particularly suited for removing fixed pattern noise in images due to image capture processes, but applies to other media signals as well. In watermarking applications, the method can be used to improve detection and recovery of a watermark. Also, it may be used to remove components of a watermark or a watermark from a cover signal that contains two or more different watermarks.

Abstract: Variable message coding protocols enable greater flexibility in encoding auxiliary data in media signals. One such protocol employs a version identifier that indicates the type of coding used to process an auxiliary data message before it is embedded in a host media signal. This version identifier specifies the type of error robustness coding applied to a variable message. The error robustness coding may be varied to alter the message payload capacity for different versions of auxiliary data embedding and reading systems. Another protocol uses control symbols to specify the format and variable length of the variable message.

Abstract: The error rate of auxiliary data embedded in media signals is decreased through variable error robustness coding. In one application, error correction coded symbols in a steganographic message that are more prone to error are repeated more than other symbols. In another application, the error robustness coding is increased or decreased in different parts of an auxiliary data message according to a measure of the expected error rate based on a model of the channel and/or the host media signal that is to carry the auxiliary data through that channel.

Abstract: Fingerprint data derived from audio or other content is used as an identifier. The fingerprint data can be derived from the content. In one embodiment, fingerprint data supplied from two or more sources is aggregated. The aggregated fingerprint data is used to define a set of audio signals. An audio signal from the set of audio signals is selected based on its probability of matching the fingerprint data. Digital watermarks can also be similarly used to define a set of audio signals.