26-02-2013, 02:37 PM
New Technique for Embedding Watermark Image into an Audio Signal
New Technique for Embedding Watermark.doc (Size: 80.5 KB / Downloads: 22)
Abstract
An audio watermarking algorithm which embedded watermark image is presented in this paper. A binary watermark image is embedded into an audio signal. For embedding process, the original audio signal is decomposed to wavelet domain by using wavelet basis. In order to obtain the robustness watermark signal, low frequency coefficients are selected and then divided to several segments. A mean of each segment is calculated and applied for the voting technique combination with mean-quantization performs embedding the watermark into the original audio signal. After original audio signal is embedded, the watermarked audio signal distortion is still inaudible. In addition, the watermark image could be recovered without using the original audio signal in extracting process. The experimental results show that the watermark image obtained from this algorithm is robust to many attack operations, such as Additive White Gaussian Noise, resampling, requantization, low pass filtering, and mp3 compression.
INTRODUCTION
The problem of copyright violation is an important problem for copyright owner. Consequently, digital watermarking techniques are occurred in pictures, movies and audios to solve this problem. An audio watermarking is also an interesting research topic. In order to protect an audio media from copying, watermark should be embedded into an original audio without any effect to the quality of original audio. Only the owner can recover this watermarking. Recently, there are several techniques which can be used to embed the watermark into an original audio file. In some literatures, the watermark could be embedded on wavelet domain [1] and spectrum domain [2]. This paper describes a new idea to improve the robust of binary image watermark embedded into audio signal. This algorithm is based on mean-quantization in DWT domain. Some parts of this algorithm presented in this study are updated from an original work [3]. However, this updated algorithm is created a new condition of voting technique. Voting technique is used to decide how we can embed the watermark. Firstly, an original audio signal is decomposed to the 4th level in DWT domain. In order to obtain the robust and inaudible watermark, low frequency coefficients are selected. Wavelet coefficients are divided to segment and mean of each segment is calculated. Next, one bit of concatenated watermark image was embedded to three neighbor segments by using voting technique and mean-quantization. Then, inverse wavelet transform is applied and watermarked audio is obtained. An extracting process becomes to reversed process of embedding process. Finally, an experiment results show that this proposed algorithm is robust to many attack operations.