Defend Against Adversarial Samples by Using Perceptual Hash

Image classifiers that based on Deep Neural Networks(DNNs)have been proved to be easily fooled by well-designed perturbations.Previous defense methods have the limitations of requiring expensive computation or reducing the accuracy of the image classifiers.In this paper,we propose a novel defense method which based on perceptual hash.Our main goal is to destroy the process of perturbations generation by comparing the similarities of images thus achieve the purpose of defense.To verify our idea,we defended against two main attack methods(a white-box attack and a black-box attack)in different DNN-based image classifiers and show that,after using our defense method,the attack-success-rate for all DNN-based image classifiers decreases significantly.More specifically,for the white-box attack,the attack-success-rate is reduced by an average of 36.3%.For the black-box attack,the average attack-success-rate of targeted attack and non-targeted attack has been reduced by 72.8%and 76.7%respectively.The proposed method is a simple and effective defense method and provides a new way to defend against adversarial samples.

Robust Watermarking Algorithm for Medical Images Based on Non-Subsampled Shearlet Transform and Schur Decomposition

With the development of digitalization in healthcare,more and more information is delivered and stored in digital form,facilitating people’s lives significantly.In the meanwhile,privacy leakage and security issues come along with it.Zero watermarking can solve this problem well.To protect the security of medical information and improve the algorithm’s robustness,this paper proposes a robust watermarking algorithm for medical images based on Non-Subsampled Shearlet Transform(NSST)and Schur decomposition.Firstly,the low-frequency subband image of the original medical image is obtained by NSST and chunked.Secondly,the Schur decomposition of low-frequency blocks to get stable values,extracting the maximum absolute value of the diagonal elements of the upper triangle matrix after the Schur decom-position of each low-frequency block and constructing the transition matrix from it.Then,the mean of the matrix is compared to each element’s value,creating a feature matrix by combining perceptual hashing,and selecting 32 bits as the feature sequence.Finally,the feature vector is exclusive OR(XOR)operated with the encrypted watermark information to get the zero watermark and complete registration with a third-party copyright certification center.Experimental data show that the Normalized Correlation(NC)values of watermarks extracted in random carrier medical images are above 0.5,with higher robustness than traditional algorithms,especially against geometric attacks and achieve watermark information invisibility without altering the carrier medical image.

A Novel Robust Watermarking Algorithm for Encrypted Medical Image Based on DTCWT-DCT and Chaotic Map

In order to solve the problem of patient information security protection in medical images,whilst also taking into consideration the unchangeable particularity of medical images to the lesion area and the need for medical images themselves to be protected,a novel robust watermarking algorithm for encrypted medical images based on dual-tree complex wavelet transform and discrete cosine transform(DTCWT-DCT)and chaotic map is proposed in this paper.First,DTCWT-DCT transformation was performed on medical images,and dot product was per-formed in relation to the transformation matrix and logistic map.Inverse transformation was undertaken to obtain encrypted medical images.Then,in the low-frequency part of the DTCWT-DCT transformation coefficient of the encrypted medical image,a set of 32 bits visual feature vectors that can effectively resist geometric attacks are found to be the feature vector of the encrypted medical image by using perceptual hashing.After that,different logistic initial values and growth parameters were set to encrypt the watermark,and zero-watermark technology was used to embed and extract the encrypted medical images by combining cryptography and third-party concepts.The proposed watermarking algorithm does not change the region of interest of medical images thus it does not affect the judgment of doctors.Additionally,the security of the algorithm is enhanced by using chaotic mapping,which is sensitive to the initial value in order to encrypt the medical image and the watermark.The simulation results show that the pro-posed algorithm has good homomorphism,which can not only protect the original medical image and the watermark information,but can also embed and extract the watermark directly in the encrypted image,eliminating the potential risk of decrypting the embedded watermark and extracting watermark.Compared with the recent related research,the proposed algorithm solves the contradiction between robustness and invisibility of the watermarking algorithm for encrypted medical images,and it has good results against both conventional attacks and geometric attacks.Under geometric attacks in particular,the proposed algorithm performs much better than existing algorithms.

Webpage Matching Based on Visual Similarity

With the rapid development of the Internet,the types of webpages are more abundant than in previous decades.However,it becomes severe that people are facing more and more significant network security risks and enormous losses caused by phishing webpages,which imitate the interface of real webpages and deceive the victims.To better identify and distinguish phishing webpages,a visual feature extraction method and a visual similarity algorithm are proposed.First,the visual feature extraction method improves the Visionbased Page Segmentation(VIPS)algorithm to extract the visual block and calculate its signature by perceptual hash technology.Second,the visual similarity algorithm presents a one-to-one correspondence based on the visual blocks’coordinates and thresholds.Then the weights are assigned according to the tree structure,and the similarity of the visual blocks is calculated on the basis of the measurement of the visual features’Hamming distance.Further,the visual similarity of webpages is generated by integrating the similarity and weight of different visual blocks.Finally,multiple pairs of phishing webpages and legitimate webpages are evaluated to verify the feasibility of the algorithm.The experimental results achieve excellent performance and demonstrate that our method can achieve 94%accuracy.