Traditional pedestrian trajectory prediction models focus on spatio–temporal data without proper consideration of individual interactions with the environment, mutual interactions, and contextual information, resulting in low prediction performance in real applications. In this article, we propose a new pedestrian trajectory prediction model called spatial social force graph neural network (SSF-GNN). First, SSF-GNN adopts a gate recurrent unit (GRU) network and a CenterNet network to capture pedestrian trajectory features and environmental features from historical trajectory sequences. Particularly, SSF-GNN can quantify pedestrian interactions and context-awareness information based on social force. Second, SSF-GNN employs a graph neural network to integrate social influence and hidden states of pedestrians. The distance between adjacent trajectory points is approximated by the weighted average summation of pedestrian historical trajectories. Third, SSF-GNN employs a new interaction function between pedestrians by considering the distance between pedestrians, as well as the movement speed of pedestrians in the social force model, to accurately predict trajectories of pedestrians. Extensive experiments are conducted on two famous datasets, and the results demonstrate SSF-GNN’s outperforms the state-of-the-art models, where average displacement error (ADE) is reduced by more than 25.6%, and final displacement error (FDE) is reduced by more than 15.4%. When predicting a pedestrian’s trajectory in the next eight frames of locations, SSF-GNN outperforms other models significantly with an accuracy of 69.71%.

Recently, deep neural networks have revolutionized the field of link prediction, and the state-of-the-art works are typically subgraph-based discriminative methods, which construct features of local subgraphs firstly and predicting potential links via deep learning based binary subgraph classification. However, the discriminative link prediction methods always fail to automatically learn features and perform link prediction, and the performance of them depends on the construction of enclosing subgraphs and the manually-designed features for the subgraphs. To address these issues, we leverage the idea of graph disentangling and propose a novel self-representation-based generative graph neural network framework (GraphLP) for end-to-end link prediction, which learns to extract the latent patterns, i.e., recurring subgraphs, from input graphs via self-supervised learning and reconstruct graphs for link prediction using the subgraphs as structural basis. GraphLP consists of three components: self-representation-based collaborative inference, high-order connectivity computation, and multi-scale pattern fusion. The key idea is to utilize the correlations between the extracted recurring subgraphs on different scales to effectively assist link inference. GraphLP also can effectively exploit the hierarchical organization patterns and incorporate them within the representation procedure, producing robust and accurate results. Compared with traditional methods and state-of-the-art methods, experimental results on public benchmark datasets demonstrate that GraphLP achieves promising performance. Different from the discriminative methods, GraphLP provides a new paradigm for generative neural-network-based link prediction.

Graph neural networks (GNNs) have been widely applied in the Internet of Things (IoT) for the intelligent analysis of data collected by sensors, particularly complex relationships and dependent information between IoT devices. However, recent studies have shown that GNNs are vulnerable to adversarial attacks, which significantly limits their application in safety-critical IoT systems such as smart health monitoring, traffic monitoring, and autonomous driving. To address this issue, in addition to the low feature similarity, this study examines the vulnerability of GNNs empirically and reveals that adversarial perturbations against GNNs tend to have low structural proximity in local neighborhoods. Thus, a natural approach for defending GNNs against adversarial attacks is to utilize the related high-order robust information of the perturbed graphs. In this study, we construct auxiliary views with high-order structure and feature similarity from a perturbed graph and propose a multi-view ensemble learning-based robust graph convolutional network (MV-RGCN). Each base model in the MV-RGCN aggregates the adversarial perturbed graph and the constructed view through an adaptive aggregation mechanism, thereby eliminating the impact of adversarial perturbations. Robust representations of the base models are then integrated using an adaptive ensemble mechanism to generate predictions. Extensive experiments under adversarial attack scenarios demonstrate that the MV-RGCN outperforms state-of-the-art methods and can achieve satisfactory performance without affecting its accuracy on the original graph data. This code is available at https://github.com/thomaslok0516/MVRGCN.

This paper presents a fresh paradigm for protecting facial privacy via an invertible image obfuscation framework that incorporates multiple characteristics including anonymity, diversity, reversibility, security, and lightweight all at once. We name the framework PRO-Face S, an acronym for Privacy-preserving Reversible Obfuscation of Face images via Secure flow. The core of the proposed framework is a flow-based generative model (or invertible neural network), which takes as input a face image along with its pre-obfuscated form, and outputs the privacy-protected image that visually mirrors the pre-obfuscated one. The pre-obfuscation applied can be in various forms with different types and strengths. The invertibility of the flow-based model ensures that the original image can be easily recovered from the protected image in high fidelity. An elaborate secret key mechanism is devised to securely guide the mutual transformations of privacy protection and image recovery, such that the correct recovery is only possible upon the availability of the correct secret, pre-specified by the user in the protection stage. Two modes of wrong recovery are investigated to deal with malicious recovery attempts in different scenarios. Finally, extensive experiments conducted on multiple image datasets demonstrate the superiority of the proposed framework over state-of-the-art methods.

Abstract The visual quality of images captured under sub-optimal lighting conditions, such as over and underexposure may benefit from improvement using fusion-based techniques. This paper presents the Caputo Differential Operator-based image fusion technique for image enhancement. To effect this enhancement, the proposed algorithm first decomposes the overexposed and underexposed images into horizontal and vertical sub-bands using Discrete Wavelet Transform (DWT). The horizontal and vertical sub-bands are then enhanced using Caputo Differential Operator (CDO) and fused by taking the average of the transformed horizontal and vertical fractional derivatives. This work introduces a fractional derivative-based edge and feature enhancement to be used in conjuction with DWT and inverse DWT (IDWT) operations. The proposed algorithm combines the salient features of overexposed and underexposed images and enhances the fused image effectively. We use the fractional derivative-based method because it restores the edge and texture information more efficiently than existing method. In addition, we have introduced a resolution enhancement operator to correct and balance the overexposed and underexposed images, together with the Caputo enhanced fused image we obtain an image with significantly deepened resolution. Finally, we introduce a novel texture enhancing and smoothing operation to yield the final image. We apply subjective and objective evaluations of the proposed algorithm in direct comparison with other existing image fusion methods. Our approach results in aesthetically subjective image enhancement, and objectively measured improvement metrics.

This paper presents the detection of brain tumors by using the VGG16 approach for grading from multiphase MRI images. It also depicts the comparative analysis among several outcomes coming from different baseline neural networks and deep learning configurations. Machine learning directly uses MRI images, with few sequential operations among multiphase MRIs. This paper illustrates the process that influences the potential of the deep learning machine. Neural networks generally involve the convolutional neural networks (CNN) for achieving the optimum enhancement on grading performance. Such processes also include visualization of kernels trained in several layers and visualize few self-learned features attained from CNN. Such research shows the deep learning approach with its applications in brain tumor segmentation. Researchers found difficulty in the automatic segmentation of brain tumors that provide great variability in sizes and shapes. Computed tomography (CT) and magnetic resonance (MR) imaging are the most widely used radiographic techniques in diagnosis, clinical studies, and treatment planning. The problems common to both CT and MR medical images are partial volume effect, different artifacts: example motion artifacts, ring artifacts, etc, and noise due to sensors and related electronic systems. In this paper, we propose an easy and unique segmentation process that provides competitive performance as well as speedy runtime for the evaluation of model performance in terms of loss and accuracy.