CAJ | 학술논문

针对现有的多功能音频水印算法恢复篡改区域的能力较弱问题，提出了一种多功能双音频水印算法。首先对原始音频信号分帧，结合Mel频谱倒谱系数计算每帧的相似帧，并将其作为鲁棒水印嵌入经过Torus映射的帧的低频小波系数的绝对平均值中；然后将高频小波系数的平均值作为初始值，利用Logistic序列生成认证水印信息，并通过单个量化高频系数的方法将其嵌入到音频信号中。仿真结果表明：所提出的算法具备很强的抗攻击性；脆弱水印对篡改敏感，而且篡改定位精确。
침대현유적다공능음빈수인산법회복찬개구역적능력교약문제，제출료일충다공능쌍음빈수인산법。수선대원시음빈신호분정，결합Mel빈보도보계수계산매정적상사정，병장기작위로봉수인감입경과Torus영사적정적저빈소파계수적절대평균치중；연후장고빈소파계수적평균치작위초시치，이용Logistic서렬생성인증수인신식，병통과단개양화고빈계수적방법장기감입도음빈신호중。방진결과표명：소제출적산법구비흔강적항공격성；취약수인대찬개민감，이차찬개정위정학。
The multifunctional audio watermarking algorithm cannot recover the tampered region. This paper designed a multifunction audio watermarking algorithm to solve the issue. First, the original audio signal was segmented into frames, and the similar frame of per frame was computed by using the Mel frequency spectrum coefficient. The similar frames as the robust watermark were embedded into the absolute mean value of audio low-frequency wavelet coefficients by using the Torus auto-morphism. Second, the average value of the high-frequency wavelet coefficients were taken as the initial value. The authentication watermark was generated by using the logistic sequence, and the watermark was embedded into the audio signal by quantifying high-frequency coefficients method. Simulation results show that the robust watermark possesses strong robustness to resist the attacks, the fragile watermark is very sensitive to tamper, and it can localize the tampered region and tamper recovery.