Research Of Image Perceptual Hash Based On Scale-invariant Feature Transform And Visual Saliency

With the rapid development of network and information technology, current society has entered the era of multimedia. Digital images with intuitive and transparent features construct a new thinking mode for human beings, which can help us to study and think. Perceptual hash technology provides a flexible effective safe solution to management of digital images through extracting a short digest of images and based on abstract matching, which becomes a new hopeful emerging technology. Image perceptual hash is widely used in the image recognition and image authentication. But at present the theoretical basis about the perception hash is not perfect. The existing algorithms are not good enough to meet the actual demand in a variety of performance. According to the deep research on the performance requirements of the two application mode, the corresponding improvement schemes is put forward in this paper to improve the effect in the practical application.(1) According to the application requirements of recognition for perceptual hash algorithm, this paper proposes a perceptual hash algorithm based on the improved Scale-invariant feature transform (SIFT). Method proposed in literature[12] uses Scale-invariant feature transform (SIFT) to extract feature points and then reduces dimensions of feature vector by PCA, which has good anti-attack ability. However, the complexity of the algorithm is too high and it is sensitive to large angle of Rotation. Firstly, SIFT is improved through reducing the layers of Gauss Pyramid, accurately calculating the main direction and reducing dimensions based on kernel projection. Secondly, perceptual features of the image is extracted by the improved SIFT. Finally, in order to increase security, use the Logistic equation as chaotic sequence generator to encrypt the perceptual features, and then quantize it to obtain the final hash.(2) According to the application requirements of authentication for perceptual hash algorithm, this paper puts forward a perceptual hash algorithm based on visual saliency. Image authentication requests perceptual hash algorithm can not only determine the authenticity of the image content but also locate the tampered area. The main strategy is to partition image into blocks and generate perceptual hash digest for each block to detect tamper. If the block is too small, the complexity of the algorithm and the size of hash will increase. If the block is too large, the accuracy of tamper detection will decline. Firstly, preprocess the original image to obtain Saliency Map and calculate adaptive sampling rate of each image block based on saliency. Secondly, sample each image block by different sampling rate. Finally, use SVD decomposition to get the final Hash. When authenticating, process the tested image with the same steps to obtain its perceptual hash sequence. First, contrast the whole hash sequence. If the authentication passes, continue to contrast the hash sequence of each block and mark the unsanctioned block as the tampered area.(3) An experiment is made to compare the method proposed for recognition with the methods proposed in the literature[12]. Experimental results show that the method proposed in this paper with lower complexity can be effectively resist the common content preserving manipulations, which is significantly better than the compared perceptual hashing algorithm. An experiment is made to compare the method proposed for authentication with several famous perceptual hashing methods. The results show that the method has a good ability of tamper detection and location under the condition of robustness.