针对特殊应用领域的立体图像安全性问题,提出了基于视点间关系预测的立体图像可逆水印,能够无损地恢复原始立体图像。根据左右视点匹配像素具有较强相关性的特征,能够较准确地预测立体图像的像素,并建立峰值点较高的预测误差直方图。通过直方图峰值点的平移,并采用多层水印嵌入,从而达到较高的水印容量。此外,根据水印嵌入代价率,为峰值寻找最优零点,从而减少了非水印嵌入的数据平移量,提高了含水印立体图像的质量。实验结果表明,相比现有预测算法,提出的基于视点间的预测算法更加精确;在相同水印容量的情况下,本文方法的图像质量高于现有扩展的立体图像可逆水印方法,在较大水印容量时,峰值信噪比(PSNR)的提高尤其明显。
To solve the security of stereo image in special applications,a stereo image reversible water-marking method using inter-view relationship prediction is proposed in this paper, which can recover the original stereo image without loss if the watermarked stereo image is not modi6ed. According to charac- teristics of stereo image, a pixel in the left view is matched with a unique pixel in the right view, and strong correlations between a pair of matched pixels are used to predict pixels accurately, so that the pre-diction error histogram with sharp peak is built. Histogram peak point shifting is used to embed water-mark,and the multi-layer embedding is utilized to increase watermarking capacity. Moreover, ratios of embedding cost are computed to search optimal zero point for the peak point, consequently data can be decreased for nor~ embedding shifting, and the quality of the watermarked stereo image is improved sig- nificantly. Experimental results demonstrate that the accuracy of the proposed inter-view relationship based predictor is better than that of the existing predictors. Furthermore,image qualities of the proposed method are better than those of the existing extended reversible watermarking methods relative to the same watermarking capacity, and PSNR value enhancement is especially obvious for large watermarking capacity.