Comparative performance of watermarking schemes using Mary modulation with binary schemes employing error correction coding

A common application of digital watermarking is to encode a small packet of information in an image, such as some form of identification that can be represented as a bit string. One class of digital watermarking techniques employs spread spectrum like methods where each bit is redundantly encoded throughout the image in order to mitigate bit errors. We typically require that all bits be recovered with high reliability to effectively read the watermark. In many watermarking applications, however, straightforward application of spread spectrum techniques is not enough for reliable watermark recovery. We therefore resort to additional techniques, such as error correction coding. As proposed by M. Kutter (1), M-ary modulation is one such technique for decreasing the probability of error in watermark recovery. It(1) was shown that M-ary modulation techniques could provide performance improvement over binary modulation, but direct comparisons to systems using error correction codes were not made. In this paper we examine the comparative performance of watermarking systems using M-ary modulation and watermarking systems using binary modulation combined with various forms of error correction. We do so in a framework that addresses both computational complexity and performance issues.