Fragile audio watermarking scheme based on sample mean sequence

摘要

In this paper, a fragile audio watermarking scheme for content authentication is proposed. The audio signal is segmented and the sum of absolute value of samples is computed for each part, then the sum for each frame is adjusted and quantized and the quantization result is converted to binary as watermark. Then, a proper positive integer is chosen to compute the mean of continuous sample amplitude value by sliding one sample every time. The mean sequence constructs a mean line which cross timing axis in time domain. Watermark information is embedded at these intersection points with dither modulation method. Experimental results show that our fragile watermarking scheme is inaudible, and it is sensitive against MP3 compression, additive noise, re-sampling, re-quantization, low-pass filtering, etc, and it can detect and localize the tampered regions effectively and accurately.