| This thesis involved the embedding of binary image as the digital watermark and its subsequent detection. The strength of this embedded watermark was controlled by a watermark scaling factor (WSF). A feedforward artificial neural network was used to determine the WSF. The WSF was properly selected for each subframe so that the power density spectrum of the watermarked signal was below the minimum masking threshold of the audio signal. Watermark embedding was done in the wavelet transform domain. The artificial Neural Network (ANN) had been used to model Human Auditory System (HAS) and the watermark had been embedded in the wavelet coefficients. This thesis involved a blind watermark detection technique so the presence of the original audio signal was not required for watermark detection. Robustness of this technique against various signal processing attacks such as jittering, cropping, low pass filtering, resampling, requantization were also studied. |
