Research On Digital Signature Techniques For Image Authentication

The authentication techniques of image authenticity and integrality are the important research subjects of the multi-media security field. In this paper, according to the requirements of image authentication, the digital signature for image integrality authentication techniques are studied based on the characteristic of digital image. The main contributions are as follows:The characters of existing typical image digital signature schemes such as robustness, sensitivity to malicious attacks, security, localization function and compactness are analyzed and evaluated. We found that perfect schemes meeting all the performances are not exist. In the practical application, an image signature scheme should be design to meet some conflicting performances by trade-offs. In addition, the performance requirements and framework of digital signature scheme for image authentication are proposed.The selection of wavelet bases for DWT-based image digital signature is discussed. The relationship between the performances of a DWT-based image digital signature and the properties of the Daubechies wavelet whose filter length is less than 20, as well as its decomposition scales are studied. According to the results of our study, the image digital signature algorithm generated by db1, db2, and db3 show stronger perceptual robustness and compactness yet lower tamper sensitivity and tamper localization capability with a higher wavelet decomposition scale. In addition, the results show that with a higher wavelet vanishing moment, the image digital signature algorithm exhibits higher computational complexity and lower compactness. The findings suggest that the db2 wavelet is suitable for the construction of the image digital signature algorithm and some rationalization proposals be proposed for the decomposition scale of the db2 wavelet.A novel digital signature scheme for image authentication is proposed. The scheme comprehensively utilize the characteristic of the DWT, DCT and KL transformation to extract the image features by two layers for reducing the signature length and concentrating the main information well. The shorter signature is obtained by quantization and Huffman coding. A mass of experimental results show that the proposed method achieves a better trade-off between robustness and the sensitivity to malicious attacks. The length of the final signature is about 450 bits and is independent of the size of the original image. Encrypting the image features extracted in the first layer, not only the key dependence and discriminative power of the signature can be obtained simultaneously, but also the security of signature and the capability of detecting tampering are enhanced. It achieves a better trade-off between security and the capability of detecting tampering with precise localization.