基于重要块二值特征的电子凭证自恢复水印算法

Self-Recovery E-evidence Image Watermarking Algorithm Based on Binary features of Important Blocks

现有高性能自恢复水印算法多是针对自然图像设计,水印容量较大且不利于文字等信息的恢复.为解决这一问题,在兼顾水印嵌入容量和篡改凭证图像恢复质量的基础上,提出了一种适用于电子凭证图像真实性认证的自恢复水印算法.该算法首先根据电子凭证图像特性,将图像块划分为重要块、非重要块和空白块3类,并对其进行分类编码;其次,为保证类型码的正确性,将所有图像块类型码置乱,采用RS编码后生成部分恢复水印信息,将其平均嵌入在所有图像块中;再次,对重要块采用二值化方法生成恢复水印信息,并依次嵌入在非重要块或空白块的最低有效位;最后,通过7张具体电子凭证图像,对比给出了本文算法与现有同类变容量文献算法的性能.研究结果表明,本文算法采用分类编码并仅对重要块生成恢复水印信息,在不降低算法篡改检测和篡改恢复性能的条件下,使水印嵌入容量从对比文献的1.73和2.99降低至0.64,含水印凭证图像的峰值信噪比分别高出对比文献8 d B和23 d B.

Existing self-recovery watermarking algorithms are generally designed for natural images, which have a large watermark embedding capacity and cannot clearly recover the text information. To deal with this, a self-recovery watermarking algorithm for E-evidence image authentication was proposed, which can reach a balance between watermark embedding capacity and recovery quality. Firstly, according to the characteristics of an E-evidence image, all image blocks in the test E- evidence image were categorized into three types： important block, unimportant one and blank one. Then, to ensure the correctness of the type code, the type code is permutated and encoded by RS coding to generate a partial-recovery watermark that is randomly embedded in all blocks. At the same time, another recovery watermark is generated by coding the binary features of important blocks and randomly embedded in the least significant bit of unimportant or blank blocks based on secret key. At last, the performances of the proposed scheme and the alterable-capacity watermarking algorithms were compared by using seven E-evidence images. Results show that the watermark embedding capacity ofthe proposed scheme is reduced to 0.64 bit per pixel, compared with two literature results of 1.73 and 2.99, and the peak signal to noise ratio （PSNR） of the watermarked E-evidence image by the proposed scheme is about 8 dB and 23 dB higher than those obtained from literature, while the performance of tamper detection and tamper recovery is not worsened. This is mainly because the proposed scheme adopts classification code and only encodes the imoortant blocks.