Search

• A unified adversarial attack framework is designed to reveal the security of the state-of-the-art image tampering localization algorithms. • Two effective attack methods are proposed by relying on the optimization-based and gradient-based adversarial example strategies. • Extensive experimental evaluations verify that the proposed attacks can sharply reduce the tampering localization accuracy while preserving high visual quality for the attacked images. It is significant to evaluate the security of existing digital image tampering localization algorithms in real-world applications. In this paper, we propose an adversarial attack scheme to reveal the reliability of such deep learning-based tampering localizers, which would be fooled and fail to predict altered regions correctly. Specifically, two practical adversarial example methods are presented in a unified attack framework. In the optimization-based adversarial attack, the victim image forgery is treated as the parameter to be optimized via Adam optimizer. In the gradient-based adversarial attack, the invisible perturbation yielded by Fast Gradient Sign Method (FGSM) is added to the tampered image along gradient ascent direction. The black-box attack is achieved by relying on the transferability of such adversarial examples to different localizers. Extensive experiments verify that our attacks can sharply reduce the tampering localization accuracy while preserving high visual quality for attacked images. Source code is available at https://github.com/multimediaFor/AttackITL. [ABSTRACT FROM AUTHOR]

In the present era of machines and edge-cutting technologies, still document frauds persist. They are done intuitively by using almost identical inks, that it becomes challenging to detect them—this demands an approach that efficiently investigates the document and leaves it intact. Hyperspectral imaging is one such a type of approach that captures the images from hundreds to thousands of spectral bands and analyzes the images through their spectral and spatial features, which is not possible by conventional imaging. Deep learning is an edge-cutting technology known for solving critical problems in various domains. Utilizing supervised learning imposes constraints on its usage in real scenarios, as the inks used in forgery are not known prior. Therefore, it is beneficial to use unsupervised learning. An unsupervised feature extraction through a Convolutional Autoencoder (CAE) followed by Logistic Regression (LR) for classification is proposed (CAE-LR). Feature extraction is evolved around spectral bands, spatial patches, and spectral-spatial patches. We inspected the impact of spectral, spatial, and spectral-spatial features by mixing inks in equal and unequal proportion using CAE-LR on the UWA writing ink hyperspectral images dataset for blue and black inks. Hyperspectral images are captured at multiple correlated spectral bands, resulting in information redundancy handled by restoring certain principal components. The proposed approach is compared with eight state-of-art approaches used by the researchers. The results depicted that by using the combination of spectral and spatial patches, the classification accuracy enhanced by 4.85% for black inks and 0.13% for blue inks compared to state-of-art results. In the present scenario, the primary area concern is to identify and detect the almost similar inks used in document forgery, are efficiently managed by the proposed approach. [ABSTRACT FROM AUTHOR]

Nowadays, various image editing tools are available that can be utilized for manipulating the original images; here copy-move forgery is most common forgery. In copy-move forgery, some part of the original image is copied and pasted into the same image at some other location. However, Artificial Intelligence (AI) based approaches can extract manipulated features easily. In this study, a deep learning-based method is proposed to classify the copy-move forged images. For classifying the forged images, a deep learning (DL) based hybrid model is presented named as VI-NET using fusion of two DL architectures, i.e., VGG16 and Inception V3. Further, output of two models is concatenated and connected with two additional convolutional layers. Cross-validation protocols, K 10 (90 % training, 10 % testing), K5 (80 % training, 20 % testing), and K2 (50 % training, 50 % testing) are applied on the COMOFOD dataset. Moreover, the performance of VI-NET is compared with transfer learning and machine learning models using evaluation metrics such as accuracy, precision, recall, F1 score, etc. Proposed hybrid model performed better than other approaches with classification accuracy of 99 ± 0.2 % in comparison to accuracy of 95 ± 4 % (Inception V3), 93 ± 5 % (MobileNet) , 59 ± 8 % (VGG16) , 60 ± 1 % (Decision tree) , 87 ± 1 % (KNN), 54 ± 1 % (Naïve Bayes) and 65 ± 1 % (random forest) under K 10 protocol. Similarly, results are evaluated based on K2 and K5 validation protocols. It is experimentally observed that the proposed model performance is better than existing standard and customized deep learning architectures. [ABSTRACT FROM AUTHOR]

In this work we have explored the hybrid deep learning architecture for recognizing the tampering from the videos. This hybrid architecture explores the features from the authentic videos to categorize the tampered portions from the forged videos. Initially, the process begins by compressing the input video using the Discrete cosine transform (DCT) based double compression approach. Then, the filtering process is carried out to improve the quality of compressed frame using the bilateral filtering. Then, the modified segmentation approach is applied to segment the frames into different regions. The features from these segmented portions are extracted and fed into hybrid DNN-AGSO (deep neural network- Adaptivf RELATED WORKSe Galactic Swarm Optimization) using Gabor wavelet transform (GWT) technique. Three different datasets are used to evaluate the overall performance they are, VTD, MFC-18, and VIRAT by MATLAB platform. The recognition rate achieved by VTD, MFC-18, and VIRAT datasets are 96%, 95.2%, and 93.47% respectively. [ABSTRACT FROM AUTHOR]