【作者】 龚素灵；

【导师】 康志伟；

【作者基本信息】 湖南大学 ， 信息与通信工程， 2010， 硕士

【摘要】 图像哈希在多媒体内容认证、数据库检索和数字水印中有着广泛的应用前景,近年来受到了高度的关注。传统的基于密码学的哈希在对图像进行压缩、过滤、几何变换等保持内容的操作时,对认证信息中每个比特的变化都非常敏感,并不适用于图像内容的认证。因此研究对内容保持操作稳健,又能够区分恶意篡改攻击的图像哈希具有重要的意义。目前使用的数字图像哈希基本上是在变换域中进行的,尤其以基于小波变换的算法居多。但是小波一般只适合于表征零维点奇异性,而图像中广泛存在的一维奇异性影响着小波特征提取的效率,同时也难以生成理想的图像哈希。有限脊波变换是在有限Radon变换域中沿每个方向作一维小波变换。它克服了小波的不足,在图像直线特征检测、图像降噪与复原等方面取得了良好的效果,但将其与图像哈希技术结合起来的研究还很少。本文在有限脊波变换应用于图像哈希算法方面作了一些研究和尝试。主要创新成果如下：(1)提出一种基于有限脊波变换与傅里叶变换结合的图像哈希算法。该算法首先对原始图像预处理,再将其分割成子块,并对每个块图像进行脊波变换,然后沿每个方向作一维傅里叶变换,最后取每个脊波块系数的最大值以及每个傅里叶变换块幅值之和作为特征向量来生成哈希序列。与未采用脊波的算法相比,该方法提高了对JPEG压缩、椒盐噪声、高斯低通滤波、旋转、缩小等操作的鲁棒性。同时,图像哈希序列由密钥控制生成,因此具有较好的安全性。(2)为了使图像哈希具有更好的鲁棒性和安全性,提出了一种新的基于视觉特性的图像哈希方法。该方法首先将原始图像预处理,然后利用人类视觉特性中‘频率敏感度提取脊波变换域上的系数生成哈希值。实验结果表明,该方法能够抵抗JPEG压缩、滤波、加噪声、剪切、旋转和缩放等攻击,具有较好的鲁棒性、敏感性和安全性。更多还原

【Abstract】 Image hash has been paid much attention in recent years for its extensive applications in content authentication, database search and watermarking. During the manipulations of content-preserving, such as compression, filtering, geometric distortion and so on, traditional cryptographic hashes are sensitive to any change of bit and not applicable to images. So it is necessary to research on the image hash to make it robust to content-preserving manipulations and able to detect the malicious attacks.Currently, image hashing algorithms are mainly implemented in transform domains. Wavelet is mostly used as one of them. However, wavelet is usually suitable for zero degree(point) singularities and the ID singularities existed in images affect the efficiency of feature extraction in the wavelet domain making it difficult to generate an ideal image hash.Finite ridgelet transform, which is 1D wavelet transform along each direction in the finite radon transform domain, has shown its advantage in linear detection and image reconstruction. However, papers in demonstrating applying finite ridgelet transform to image hashing algorithm are few. In this dissertation, the author has made some research and attempt to do so. The main innovations are as follows:(1)An image hashing based on finite ridgelet transform and fourier transform is proposed. At first, it preprocessed the original image and partitioned the image into blocks. Second, it made ridgelet transform for each image block and then made 1D FFT transform along each direction in the ridgelet transform domain. Finally, it extracted the maximum value of ridgelet coefficients and the sum of FFT amplitudes to generate the hash value. Compared with some other hashing methods, the ridgelet-based hash has better robustness against JPEG compression, salt & pepper noise, gaussian lowpass filtering, rotation, scale and so on. Since a key is used in the algorithm, the hash is harder to be forged.(2)To obtain more robust and secure image hash value, a novel image hashing scheme based on human visual system is proposed. It first preprocessed the original image, then extracted ridgelet coefficients based on frequency sensitivity in the human visual system to generate the hash value. The experimental results demonstrate that the scheme has a good performance against JPEG compression, filtering, adding noise, shearing, rotation and scale. The hash has good robustness, sensitivity and security.更多还原