针对立体图像内容的真实性认证和完整性保护等问题,提出一种基于双目恰可觉察失真(BJND)的立体图像篡改定位及恢复水印方法。首先,利用奇异值的稳定特性,设计左右图像的定位水印;然后,根据BJND强度将左右图像的定位水印嵌入到左图像中;最后,利用离散余弦变换(DCT)和JPEG量化压缩生成恢复信息,并将右图像和左图像遮挡暴露区域的恢复信息分别嵌入到右图像和左图像中。实验结果表明,本文方法所构造的水印对JPEG压缩、椒盐噪声、高斯白噪声等偶然攻击比较鲁棒,而对剪切、拼接、旋转等恶意攻击较为脆弱,同时对左右图像的恶意篡改区域检测率大于98%。由于充分利用了左右图像的匹配特性,恢复的被篡改立体图像其左右图像PSNR达36.02dB~41.29dB。
To address issues of content authenticity verification and integrity protection, a watermarking method based on binocular just-noticeable-difference (BJND) is proposed for tamper localization and recovery of stereoscopic images. Firstly, for its strong stability, singular values of the left and right images of a stereoscopic image are used to construct authentication-bits for the left and right images. Then, the generated stereoscopic image authentication-bits are embedded into the left image by making use of the binocular just-noticeable-different value. Finally, the recovery information of right image and occlusion parts in the left image is produced with discrete cosine transform and JPEG quantization compression, and embedded into the right and left images themselves, respectively. The experimental results show that the watermarking constructed with the proposed method isrobust to incident attacks, such as JPEG compression, salt and peppy noise and Gaussian white noise, but relatively fragile to mailicious attacks, for example, crop, forgery, rotate, etc. The accuracy of tamper detection is higher than 98~ for malicious attacks. For different stereoscopic images degraded by malicious attakcs, the peak signal-to-noise ratios (PSNRs) of the recovered left and right images of the stereoscopic images are from 36.02 dB to 41.29 dB,by taking full advantage of the matching characteristics between the left and right images.