Image Authentication Algorithm Based On Momentinvariant Of Histogram And SVD

For multimedia data, authenticity is very important, and usually we need to verify the authenticity (also known as "integrity") of image. The three methods, which are based on digital watermarking, mentioned in the thesis are robust image authentication method combing SVD and PKI, image authentication method based on JPEG invariant and digital watermarking, and image authentication algorithm based on moment invariant of histogram and SVD. Through the function analysis, we can get the advantages and disadvantages of all the three methods above, then the result is that the image authentication algorithm based on moment invariant of histogram and SVD is the one that suits for all performance index of image authentication system. The thesis represents some further research results about the image authentication algorithm based on moment invariant of histogram and SVD, meanwhile, there are also some improvements on the method.Based on the good invariance of the image's moment invariant of histogram, we proposed a new method. The new method used moment invariant of histogram as the watermark, and defined the invariance by the information of moment invariant of histogram. SVD in numerical analysis was the numerical algorithm that diagonalize the matrice. The operation and inverse operation of SVD have finished the embedding and extraction of watermarking, which is for the image authentication.First, we need to partition the image, extract the moment invariants of histogram one by one, and combine them as the watermark. The next is to embed the watermark into the original image in the way of SVD. The last is to judge if the image is distorted and the distorted location by the changed value Y and relative coefficient.The thesis also represents the theoretical analysis on the method's resistance to geometrical attacks, which has been approved in the experiment of Maglab7.0. During the experiment, we distorted the image with watermarking, contrasted the image with watermarking and the original image, and judged the distortions and exact locations, then attacked the distorted image geometrically, such as transposition, mirrorimage, and enlargement. In the end, the distortions were discovered by value Y and some relative coefficient.The result of this experiment shows that this kind of algorithm applies to detecting BMP image distortion, and can accurately determine the location of the image distortion. Plus, this algorithm can be strongly resistant to geometric attack.