This paper reports comparative evaluations of the watermarking methods we previously proposed with four other methods. Three of the proposed methods are based on the concept of embedding inaudible watermarks into an original sound by controlling the phase characteristics of the sound in relation to the characteristics of cochlear delay (non-blind, blind, and reversal CD methods; Unoki et al., 2008; 2011). The other one is based on formant enhancement by embedding watermarks into line spectral frequencies (LSFs) that obtained from linear prediction (LP) (LSFs-LP method; Wang and Unoki, 2014). Four other methods are typical methods of least significant bit-replacement (LSB) method, direct spread of spectrum (DSS) method, Echo-hiding (ECHO) method, and singular value decomposition (SVD)-based method. We experimentally evaluated the proposed methods with the other methods by carrying out four objective tests (PEAQ, LSD, SNR, and WSS), bit-detection test, and robustness tests (against down sampling, requantization, MP3 and MP4 compressions, bandpass filtering, Gaussian noise addition, pitch change, a single echo addition, and a series of attacks that mimic DA and AD conversions). Most of the evaluations we used are recommended by the IHC (Information Hiding and its Criteria for evaluation). From the results of comparative evaluations, we found that although LSB had a drawback in robustness for watermarking, it could satisfy inaudibility. We also found that DSS could satisfy robustness, but DSS had a drawback with inaudibility. For the other methods of ECHO and SVD, there was a trade-off between inaudibility and robustness requirements. The results revealed that our proposed methods (CD methods) could embed inaudible watermarks into original sounds and could precisely and robustly detect the embedded data from the watermarked sounds. These comparative results also suggested how our proposed methods should be developed to be useful way of information hiding in the future Internet.
展开▼