Improved digital watermarking schemes using DCT and neural techniques

Abstract

The present thesis investigates the copyright protection by utilizing the digital watermarking of images. The basic spatial domain technique DCT based frequency based technique were studied and simulated. Most recently used Neural Network based DCT Scheme is also studied and simulated. The earlier used Back Propagation Network (BPN) is replaced by Radial Basis Function Neural Network (RBFNN) in the proposed scheme to improve the robustness and overall computation requirements. Since RBFNN requires less number of weights during training, the memory requirement is also less as compared to BPN.
Keywords : Digital Watermarking, Back Propagation Network (BPN), Hash Function, Radial Basis Function Neural Network (RBFNN), and Discrete Cosine Transform (DCT). Watermarking can be considered as a special technique of steganography where one message is embedded in another and the two messages are related to each other in some way. The most common examples of watermarking are the presence of specific patterns in currency notes, which are visible only when the note is held to light, and logos in the background of printed text documents. The watermarking techniques prevent forgery and unauthorized replication of physical objects. In digital watermarking a low-energy signal is imperceptibly embedded in another signal. The low-energy signal is called the watermark and it depicts some metadata, like security or rights information about the main signal. The main signal in which the watermark is embedded is referred to as the cover signal since it covers the watermark. In recent years the ease with which perfect copies can be made has lead large-scale unauthorized copying, which is a great concern to the music, film, book and software publishing industries. Because of this concern over copyright issues, a number of technologies are being developed to protect against illegal copying. One of these technologies is the use of digital watermarks. Watermarking embeds an ownership signal directly into the data. In this way, the signal is always present with the data.
Analysis
Digital watermarking techniques were implemented in the frequency domain using Discrete Cosine Transform (DCT). The DCT transforms a signal or image from the spatial domain to the frequency domain. Also digital watermarking was implemented using Neural Networks such as:
1. Back Propagation Network (BPN)
2. Radial Basis Function Neural Network (RBFNN)
Digital watermarking using RBFNN was proposed which improves both security and robustness of the image. It is based on the Cover’s theorem which states that nonlinearly separable patterns can be separated linearly if the pattern is cast nonlinearly into a higher dimensional space. RBFNN contains an input layer, a hidden layer with nonlinear activation functions and an output layer with linear activation functions.
Results
The following results were obtained:-
1. The DCT based method is more robust than that of the LSB based method in the tested possible attacks. DCT method can achieve the following two goals: The first is that illegal users do not know the location of the embedded watermark in the image. The second is that a legal user can retrieve the embedded watermark from the altered image.
2. The RBFNN network is easier to train than the BPN network. The main advantage of the RBFNN over the BPN is the reduced computational cost in the training stage, while maintaining a good performance of approximation. Also less number of weights are required to be stored or less memory requirements for the verification and testing in a later stage.