Independent Component Analysis Applied to Watermark Extraction and its Implemented Model on FPGAs

Abstract

Most of published audio watermark algorithms are suffered a trade-off between inaudibility and detectibility, and the detection performance depends greatly on the strength of noise added by communication channels. This work introduces an audio watermarking method that can overcome this challenge, i.e. allows increasing watermark strength while preserving inaudibility. The scheme uses psychoacoustic masking compatible to MPEG layer 1 Model 1 and adjusts it in a data adaptive way. A blind watermark extraction technique using the Independent Component Analysis (ICA) is shown to minimize the watermark decoding error. An implementation of a simple quantization-based watermarking algorithm (LSB) on the Spartan-3 FPGA Starter Kit of Xilinx is also presented as a part of hardware demonstration of the method.