Image Hashing Algorithms Based On Invariant Moment And Compressed Sensing

The extensive application of image processing software makes a large number of images with similar visual content appear on the Internet.How to efficiently detect the visual similar images of a given image is an important problem to be solved in the field of image processing.In recent years,more and more researchers have proposed to use image hash algorithm to solve this problem.The image hash algorithm can map an image into a short sequence of numbers.This sequence of numbers is called an image hash,which is an image representation based on visual content and does not depend on the specific bit data of the saved image.Because the data of the hash sequence is simple and its length is short,the storage cost of storing the hash sequence and the speed of calculating the hash similarity are both low.Using the hash sequence instead of the image itself can realize the efficient processing of the image.Currently,image hashing algorithms have been applied to many aspects of image processing,including retrieval,authentication,and quality evaluation.In general,the design of hash algorithms for digital images needs to meet the two basic performance indicators of robustness and uniqueness.Robustness is mainly for images with the same visual content,regardless of whether their resolution and the specific stored bit data are the same,the hash algorithm should calculate the same hash sequence or similar hash sequence.The uniqueness is mainly for images with different visual contents,and the hash algorithm is required to map them into hash sequences with different values.Designing a hash algorithm that takes into account the above two performance indicators is an important task in current research.In this paper,we use the theory and technology of invariant moments and compressed sensing to study new image hashing algorithms,and design two new image hashing algorithms.The first algorithm is a hash algorithm based on visual saliency model and invariant moments.This algorithm uses the visual saliency model to ensure robustness,and achieves distinguishing performance through invariant moment features.The second algorithm is a hash algorithm based on compressed sensing and ordinal measurement.The algorithm uses compressed sensing theory to achieve data compression,and uses ordinal measurement to construct a quantized robust hash.The main research results are summarized as follows.1.Image hashing algorithm based on visual saliency model and invariant moment is proposed.The visual saliency model can effectively detect the visually saliency area of the image without changing the moment and has good robustness and distinguishability.To this end,the visual saliency model and invariant moments are used together to study the image hash algorithm,and a new hash algorithm is designed.An important contribution of the algorithm is to construct a weighted image representation using discrete wavelet transform low-frequency subbands and visual saliency models.The use of visual saliency model ensures the robustness of the algorithm.In addition,since the invariant moment has good robustness and distinguishability,extracting the invariant moment from the weighted image representation to construct a hash sequence can ensure that the algorithm has good classification performance between robustness and uniqueness.A large number of experiments are used to verify the performance of the algorithm,and the results show that the hash algorithm has good robustness and uniqueness.The receiver operating characteristic curve is selected as a tool for analyzing the comparison results.The experimental results show that the hash algorithm is superior to various literature hash algorithms in the classification performance between robustness and uniqueness.2.Image hashing algorithm based on compressed sensing and ordinal measure is designed.Compressed sensing is a new signal processing theory emerging in recent years,which can achieve efficient compression of data.Considering that the nature of image hashing is also a data compression technology,a new image hashing algorithm is designed using compressed sensing theory and ordinal number measurement technology.The algorithm first uses the Itti visual saliency model and Canny operator to construct a weighted image representation,then uses compressed sensing to extract compact features from the weighted image representation,and finally uses the ordinal measure of compressed sensing features to construct a hash value.The hash similarity judgment is realized by calculating the L2 norm.Various experimental tests were conducted on the algorithm,including robustness verification,uniqueness test,block size selection,visual saliency model selection,quantization scheme selection,and ordinal measure validity verification.The experimental results verified the performance of the algorithm.Compared with the performance of various literature hash algorithms,the results show that the classification performance of the algorithm on the receiver operating characteristic curve is better than the comparison of the literature algorithms.