A NEW STATISTICAL DETECTOR FOR DWT-BASED ADDITIVE IMAGE WATERMARKING USING THE GAUSS–HERMITE EXPANSION

INTRODUCTION

A digital watermark is a distinguishing piece of information that is embedded into the data with the intention of protecting the intellectual property rights, verifying the authenticity of the data, annotating the data, etc. Very often, a watermarking scheme is seen as a secured communication task consisting of two steps, watermark embedding and watermark retrieval. In watermark embedding, the signal, i.e., the watermark, is transmitted through the host data that acts as a channel, whereas in watermark retrieval, the signal is received and extracted from the marked data. The security of the watermarking is maintained in the same way as in cryptography by using a secret key. In contrast to encrypted data, a watermarked data can still be used with the embedded watermark. The legal owner knows the exact embedding process, which is based on a secret key, and, hence, can extract the watermark, while it is not possible do so for an unauthorized party. Thus, watermarking does not necessarily prevent the copying of the marked data; rather, the scheme identifies the original data source and its intended legitimate destination, so that copyright violations can at least be detected. 

Digital image watermarking can be of two types: imperceptible and perceptible. Very often, the imperceptible watermarking is preferred to the perceptible one, since the distortion of the original image is minimum in the former than that in the latter, even in the case of embedding multiple watermarks. The imperceptibility constraint in the image watermarking is achieved by taking into account the properties of the human visual system (HVS) during the embedding process. There are several methods of watermark embedding, such as additive, multiplicative, and through quantization. Whatever be the embedding process, the watermark travels with the marked image and, therefore, remains protected until it is removed.

In order to resolve the rightful ownership of an image, detection of the watermark is sufficient assuming that the owner of the marked image possesses the exact secret key and has the knowledge of the embedding process. If a marked image has a single watermark, the detection of the watermark becomes simple provided the original host image is available to the watermark detector. But, for a marked image having multiple watermarks there is ambiguity in resolving the rightful ownership, even when original host image is available to the watermark detector. In addition to this fact, in many applications such as data monitoring and tracking on the Internet, the original host data is not always available and in many multimedia settings such as video watermarking, the use of unmarked host data is not practical due to its huge volume. As a result, it is necessary to design a blind watermark detector that does not require the original host data in the detection process.