The exponential growth in PVT corners due to Moore's law scaling, and the increasing demand for consumer applications and longer battery life in mobile devices, has ushered in significant cost and power-related challenges for designing and productizing mobile chips within a predictable schedule. Two main reasons for this are the reliance on human decision-making to achieve the desired performance within the target area and power budget, and significant increases in complexity of the human decision-making space. The problem is that to-date human design experience has not been replaced by design automation tools, and tasks requiring experience of past designs are still being performed manually.

We propose a new paradigm for blind watermark decoding in the presence of desynchronization attacks. Employing Forney-style factor graphs to model the watermarking system, we cast the blind watermark decoding problem as a probabilistic inference problem on a graph, and solve it via message-passing. We study a wide range of moderate to strong attacks including scaling, amplitude modulation, fractional shift, arbitrary linear and shift-invariant filtering, and blockwise filtering, and show that the graph-based iterative decoders perform almost as well as if they had exact knowledge of the desynchronization attack parameters. Other desirable features of the graph-based decoders include the flexibility to adapt to other types of attacks and the ability to cope with the “curse of dimensionality” problem that seemingly results when the desynchronization parameter space has high dimensionality. These properties are unlike most blind watermark decoders proposed to date.

We address the question of constructing practical blind watermark decoders that are resilient to scaling and amplitude modulation attacks. We cast the watermark decoding problem as a probabilistic inference problem on a factor graph and solve it via message passing. Our decoder performs nearly as well as a decoder that has exact knowledge of the desynchronization parameters. Experimental results, on both synthetic and natural images, suggest that the decoding algorithm is capable of handling moderate to strong desynchronization attacks very well.

In this paper, we study some fundamental performance limits of blind data hiding against desynchronization attacks. These attacks are modeled in addition to independent Gaussian noise to the marked signal, followed by linear, time-invariant filtering. We study a joint estimator-decoder which estimates the desynchronization attack parameters and uses these estimates in the decoding step. We propose a coding scheme based on distortion-compensated quantization index modulation and derive the estimation accuracy of the attack parameters via Fisher information and a Cramer-Rao type bound. For illustration purposes, we report estimation and decoding results on several attacks, including classical ones (scaling and fractional shifts) and some new ones. The results are in close agreement with our bounds and tightly quantify the performance loss due to desynchronization and the influence of code block length. Thus our results demonstrate the high performance of a joint estimation-decoding approach.

The performance of current blind watermark decoders against desynchronization attacks is rather poor. Such attacks include filtering, amplitude modulation, gamma correction, time-varying delays, and spatial warping. We propose a new family of watermark decoders based on modern methods for iterative decoding using graphical models. This approach addresses the "curse of dimensionality" problem that seemingly results when the desynchronization parameter space has high dimensionality.