Implementation Of Watermarking Technique Using NVIDIA GPUs

Abstract

The proliferation of digitized media due to the rapid growth of networked multimedia systems, has created an urgent need for fast copyright enforcement technologies that can protect copyright ownership of multimedia objects. Digital image watermarking is one such technology that has been developed to protect digital images from illegal manipulations. In particular, watermarking algorithms which are based on the discrete wavelet transform have been widely recognized due to its excellent spatial localization, frequency spread, and multi-resolution characteristics, which are similar to the theoretical models of the human visual system. In this project,an imperceptible and a robust DWT based watermarking algorithm have been develop using MATLAB and CUDA - Compute Unified Device Architecture. MATLAB is a very powerful tool which deals with immensely large image manipulations. There are several ways available to write code in MATLAB for large calculations but they are very time consuming. As a solution, the algorithm is implemented in MEX and CUDA MEX, which takes less time compared to M-files implementation. The comparison results with respect to time in between the M-files, MEX-files, CUDA MEX file has been measured. The parallel CUDA MEX code will run in many threads on an NVIDIA graphical processing unit. By decreasing the completion time of the longest part of the code, largest performance increase can be achieved.