ISCTIS 2022 Keynote Speakers
Speech Title:SGCN: Spatially Gradient Convolution Network for Certificate Document Image Manipulation Localization
Abstract: Current tampering detection methods pay more attention to natural content images. The research on tampering algorithms for certificate document images is relatively limited, but certificate document images are the most commonly tampered images, and they cause great harm to society. In this work, we propose a network ASGC-Net for certificate document-like image manipulation detection based on the spatial attention mechanism. To achieve a network that can better localize text tampering cues, we also propose a novel spatially constrained convolution that can effectively suppress image content and adaptively learn manipulation detection features by capturing the difference features between the neighborhood and the center of the convolution space. To increase the network's ability to capture tampering cues at multiple scales of images, we add multilayer cross-scale connections inspired by FPN networks. Experimental results show that the algorithm can locate the tampered regions of certificate document images more accurately than general-purpose manipulation detection algorithms.
Speech Title: On virtual big data, with an application to parameter estimation problem
Abstract: Big data is significant for a variety of science and technology problems. Nevertheless, big data is hard to obtain except some special areas. Different from practical activities, Monte Carlo sampling can easily generate a great number of sample data (observations) from a population of random variable. This presentation addresses the application of big data in establishing criterion for parameter estimation. Based on a sample of a Weibull distribution with known parameters, the estimates of the parameters can be obtained by an appropriate parameter estimation method, and the errors of the estimates can be evaluated easily by comparing with the true values. As different criteria underlain the parameter estimation method subject to different errors, the criteria can be evaluated based on the parameter estimation results from a large number of samples. In the situation of small size of samples, it is demonstrated that such established criterion yields more accurate, more robust parameter estimates than theoretical criterion does, since the theoretical criterion suffers from sample uncertainty effect more.
Speech Title: Normalized Laplacian Spectrum and Internet topology
Abstract: Compressing network scale is of great significance for reducing the cost of network simulation and network visualization. This report introduces our recent research on the normalized Laplacian spectrum and Internet topology. Specifically, we found that the spectrum is scale-independent and indicates many structural characteristics of the topology. In addition, we designed a graph sampling algorithm using the spectral features, which can reduce the scale of the Internet topology by more than 96% while maintaining important graph properties.