基于多码本矢量量化的图像篡改恢复

提出一种用于数字图像篡改检测与内容恢复的自嵌入脆弱图像水印算法.对原始图像每个不重叠分块进行基于边缘复杂度的内容分析,选择不同大小的矢量量化码本进行分块压缩,可获得相应的标识符和表示图像分块主要内容的索引值以构成压缩码.将其复制多份后作为参考比特,通过密钥嵌入至其他多个分块中.用于篡改定位的认证比特则被嵌入每个分块本身.由于在多码本矢量量化协同下生成的参考比特具有较高的内容表示效率,该算法在篡改率相同的条件下可取得更理想的内容恢复性能,这一结论得到了实验结果的验证.

Image Tampering Recovery Based on Multi-codebook Vector Quantization

We propose a self-embedding fragile watermarking scheme to detect image tampering and recover original contents. By analyzing edge complexity of all non-overlapping blocks in the original image, the codebooks of vector quantization with different sizes are chosen to compress these blocks. The corresponding indicator and the index value representing the main content of the image block are collected to form the compression codes. They are duplicated to be used as reference-bits. These reference bits are then embedded into other multiple blocks according to secret keys, and authentication-bits are embedded into each block itself. Since the reference-bits based on multiple-codebook vector quantization have high efficiency in representing the image contents, the proposed scheme can achieve good performance of content recovery as compared to the reported methods. This is verified by experimental results.