Perceptual Image Hashing Based On Information Fusion And Sift

With the rapid development of information technology, digital image got a rapid expansion in the number and constant innovation in the application. Therefore, but it's a looming problem that how to certificate the authenticity of the digital images and how to manage a large number of digital images. n image hash function maps an image to a short binary string based on the image's appearance to the human eye, and it can reflect the image well. So perceptual hash got a widely attention and was considered a promising emerging technologies. For cryptographic hash functions such as the well known MD5 and SHA1, any small change in the original message, even one bit, will significantly change the hash value. This makes them unsuitable for image as an image may undergo various digital manipulations, e.g., filtering, adding noise, compression coding, and geometric modification. These normal operations do not alter the essential content of the image, therefore should not significantly change the hash value. So the image hash values should depend on image contents. So it has a important research that how to keep operating robustness in the content keeping, and more sensitive to malicious tampering.In this thesis we study the perceptual image hash theoretical analysis and specific algorithm design based on the image contents authentication. We indicate two perceptual image hash algorithms which compromise well of the only one sex, robust, and security of the image contents.The main contents of this thesis are as follows:(1)Give a perceptual image hashing algorithm based on non-negative matrix factorization (NMF) algorithm and information matched layer fusion of Hu invariant moments. It fusion the NMF hashing and Hu hashing in the matched layer. Compared with other three classical algorithms, the experimental results indicate that this algorithm is robust to the image compression, filtering, add noise and Geometric transform and so on and this algorithm is sensitive to the malicious tamper of images. The generating of the hash sequence is based on key in this algorithm, ensuring the security of the algorithm. The retrieval experimental and the auihcmication experiment of the algorithm prove that this algorithm has very good receiver operating characteristies and retrieval performance. Compored with other three algorithm, the robustness of this algorithm with the operating characteristics of the receiver were improved.(2) In order to overcome the existing perception hash algorithm which can't resist the attacks of large angle rotation attack, sunshine and so on, an perceptual image hashing algorithm based on SIFT is proposed. This algorithm used the improved SIFT method extract the image feature vector and project the feature vector to the pseudo-random vectors which is generated based on key control, this get hash sequence. Experimental results prove that the method can not only against JPEG compression, filtering and so on that common content keeping manipulations, still can resist the large angle rotation attack. And the experimental results also indicate that this algorithm also sensitive to malicious tampering, and has a good compromise to the uniqueness of the image content, robustness and security.