Research On Image-aware Hashing Technology Based On Approximately Invariant Feature

With the widespread popularity of smartphones and the rapid development of mobile internet,people are gradually shifting from traditional text-based communication to the transmission of images and videos.While this transition enriches information exchange,it also brings forth a series of issues.For instance,images consume more storage space.Furthermore,with the gradual rise of image editing software,image tampering detection has become one of the hot topics in network security research.To address these problems,researchers have proposed the technique of perceptual image hashing,which maps the content of an image to a digest value associated with its content.This technology has been widely applied in various fields such as image retrieval,image authentication,tampering detection,and quality assessment.This article expounds on the research background and significance of perceptual image hashing technology,as well as the fundamental knowledge in the field of perceptual image hashing,including the properties and framework of perceptual image hashing.By classifying the current state of perceptual image hashing research,it is found that there are some limitations in existing perceptual image hashing techniques.To address these limitations,two new perceptual image hashing algorithms are proposed in this article.One is based on the reconstruction histogram,and the other is based on deep learning.The main findings of this article are summarized as follows.1.An image perceptual hashing algorithm based on reconstructed histograms is proposed.Histogram is a feature commonly used in the field of image perceptual hashing.Due to the shape invariance of histograms,most image perceptual hashing algorithms based on histograms are very robust against geometric attacks.However,these algorithms cannot resist attacks that only change the position of pixels without changing their values,such as Arnold permutation attack.To solve this problem,we reconstruct the histogram using the relationships between pixels and their surrounding pixels,and extract the hash value from the reconstructed histogram.Through several experiments,our algorithm has good robustness and discrimination.In particular,our algorithm can resist rotation attacks from any angle.By using receiver operation characteristic curve analysis,our algorithm performs better than several algorithms proposed in the literature.2.An image perceptual hashing algorithm based on HED and principal component analysis is proposed.Image edge is a very important image feature in the field of image processing,and image edge contains rich information.Most of the existing image perceptual hashing techniques based on image edges use traditional edge detection algorithms to extract image edges,such as the Canny operator.However,traditional edge detection techniques only focus on local changes in the image and do not understand the semantics of the image,which can affect the effectiveness of edge detection.To address this issue,we use Holistically Nested Edge Detection(HED)to extract image edges,HED is a deep learning model.Then,we use principal component analysis technology to extract the final hash value from the extracted edges.Experimental verification shows that the algorithm is robust to common content retention operations,such as brightness and contrast adjustments.It also has good distinguishability.Through comparative experimental analysis,our algorithm performs better than several algorithms proposed in the literature.