DCT-DWT Based Digital Watermarking and Extraction using Neural Networks

Recent trends in information technology have miraculously changed the way people interact with multimedia devices. The steady and consistent developments in communication technologies have created a situation where a massive amount of data and thereby information is created, stored and shared every second. With an increased number of devices, backed by enhanced communication technologies the insurgence towards originality and ownership of the data is increasing. Among various other methods of protecting the originality and ownership of the digital data, watermarking has emerged to be a prominent one. Watermarking allows the copyrighting of the data through algorithm-driven encoding techniques. Along with imperceptibility and robustness, digital watermarking also requires default requirements such as security, fidelity, inseparability, payload capacity and effectiveness. For a digital watermark to be inseparable, the carrier signal(secret key) should remain undistorted outside certain specific conditions. The aim here is to create an algorithm which can efficiently create strong a digital watermark and which can also extract the watermark without the need of the original image using neural networks. Here we propose a neural network-driven digital watermarking with functions in frequency domain transformation. A strong DCT-DWT hybrid algorithm is used to embed a visible watermark into the original image and at later stages neural networks are deployed to extract the watermark without the use of the original pure image.