Image Hashing Algorithms Based On Tensor Decomposition

The popularity of mobile devices such as smartphones and tablet computers has made it increasingly convenient to obtain digital images,and the popularity of online social platforms has led to an explosive growth of digital images on the Internet.In the face of massive image data,how to store and manage them is an important problem faced by current research.Image hashing algorithm is a cross research topic in the field of digital image processing and information security,and it is an effective technology to achieve efficient management of massive images.It extracts a compact sequence of numbers based on the visual content of the image,that is image hashing.Since the length of the image hashing is short,using the image hashing to represent the image itself can not only reduce the storage space of the image,but also simplify the similarity calculation of the image.Generally,an image hashing algorithm should have two most basic performance indicators,namely robustness and discrimination.Robustness requires a hashing algorithm to map images with similar visual content to the same or similar hashes,while discrimination requires a hashing algorithm to map images with different visual content to different hashes.There are constraints on the basic performance indicators of image hashing algorithms.Designing new hashing algorithms to simultaneously improve these two performance indicators is an important task of current research and is of great significance.Tensors are generalized representations of higher-order matrices.Tensor decomposition can decompose a tensor into multiple low-dimensional matrices,which is a useful data analysis technique.Generally,there are two classical decomposition techniques for tensor decomposition,namely CP decomposition and Tucker decomposition.In order to improve the classification performance of hashing algorithms in terms of robustness and discrimination,this dissertation uses CP decomposition and Tucker decomposition,combined with discrete cosine transform and visual saliency model to carry out research on image hashing algorithms,and designs two new image hashing algorithms based on tensor decomposition.The first is image hashing algorithm based on CP decomposition and discrete cosine transform,and the second is image hashing algorithm based on Tucker decomposition and visual saliency model.The main findings of this dissertation are as follows:1.Image hashing algorithm based on CP decomposition and discrete cosine transform is proposed.CP decomposition is a useful data analysis technique that decomposes a third-order tensor into three factor matrices.In this dissertation,an efficient image hashing algorithm is designed using CP decomposition and discrete cosine transform.The specific steps are as follows.First,the input image is preprocessed to generate a normalized image.then the discrete cosine transform is used to process the three color-components of the color image,and the low-frequency coefficients are selected to generate three feature matrices.Since the low-frequency coefficients contain most of the energy information of the image and are less disturbed by noise,selecting the low-frequency coefficients to construct the matrix can ensure discrimination and robust performance.Next,the feature matrix is non-overlapping blocks,and a third-order tensor is constructed by stacking blocks.Finally,the CP decomposition is applied to the third-order tensor,and the row mean of each factor matrix is calculated to obtain the corresponding eigenvector,and the mean quantization of the eigenvector is performed to generate a hash sequence.Since the factor matrix effectively preserves the topology of the original tensor,generating hashes with the factor matrix ensures better discrimination.Using two public image databases to verify the performance of the algorithm,the experimental results show that the hash algorithm can resist a variety of digital operations,and the classification performance is better than many existing image hashing algorithms.2.Image hashing algorithm based on Tucker decomposition and visual saliency model is proposed.Tucker decomposition is another commonly used tensor decomposition technique that decomposes a third-order tensor into three factor matrices and a core tensor.The visual saliency model can detect the salient regions of the image that are of interest to the human visual system,and the use of the image saliency map to calculate the hash can improve the robust performance.Therefore,this dissertation combines Tucker decomposition and visual saliency model to carry out research on image hashing algorithm,and designs a new hashing algorithm.The specific steps are as follows.First preprocess the input image;then use the LC visual saliency model to extract the saliency map.Then,non-overlapping blocks are performed on the saliency map,and a third-order tensor is constructed by stacking image blocks.Finally,the Tucker decomposition is applied to the tensor,and the first column vector of the three factor matrices is selected to construct the intermediate hash sequence,The intermediate hash sequence is encrypted and quantized using a piecewise linear iterative chaotic map to generate the final binary hash.Extensive experiments are conducted on two public image libraries,and the results show that the algorithm has better discrimination and security.Compared with a variety of document hashing algorithms,the results show that the algorithm has certain advantages in classification performance.