Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

This paper proposes an audio tampering detection method based on the ENF phase and BI-LSTM network from the perspective of temporal feature representation learning. First, the ENF phase is obtained by discrete Fourier transform of ENF component in audio. Second, the ENF phase is divided into frames to obtain ENF phase sequence characterization, and each frame is represented as the change information of the ENF phase in a period. Then, the BI-LSTM neural network is used to train and output the state of each time step, and the difference information between real audio and tampered audio is obtained. Finally, these differences were fitted and dimensionally reduced by the fully connected network and classified by the Softmax classifier. Experimental results show that the performance of this method is better than the state-of-the-art approaches.

Audio Tampering Forensics Based on Representation Learning of ENF Phase Sequence

A digitally altered photograph, often leaving no visual clues of having been tampered with, can be indistinguishable from an authentic photograph. As a result, photographs no longer hold the unique stature as a definitive recording of events. We describe several statistical techniques for detecting traces of digital tampering in the absence of any digital watermark or signature. In particular, we quantify statistical correlations that result from specific forms of digital tampering, and devise detection schemes to reveal these correlations.

Statistical tools for digital forensics

Model-data fusion of hydrologic simulations and GRACE terrestrial water storage observations to estimate changes in water table depth

\n Hydrocarbon fields that are located offshore Abu Dhabi, United Arab Emirates (UAE), are known to be associated with undulating thick sedimentary sequences. These undulations are mostly influenced by variations in the depth of Infracambrian Hormuz salts that generate negative gravity anomalies. Nonetheless, a few known oil fields are uncorrelated with the airborne gravity observations. This is attributed to the interference from large positive gravity anomalies from basement highs. To filter out the effect of basement, we calculate the pseudogravity effect of the airborne magnetic anomalies and subtract it from the gravity anomalies. The resultant gravity anomalies mainly represent the effect of the salt domes. The results uncover deep salt structures and introduce potential traps for hydrocarbons that have proved difficult to map accurately with current seismic techniques. A nonlinear 3D inversion modeling of corrected magnetic and decreased gravity data is also used to determine the depth to basement and the Infracambrian Hormuz salts over two regions. Our findings demonstrate that the depth to basement in these regions changes from 7100 to 9700 m, and the depth to Infracambrian Hormuz salt changes from 5800 to 9400 m, with a variable thickness with a maximum of 2700 m.

Determining Infracambrian Hormuz Salt and Basement Structures Offshore Abu Dhabi by Joint Analysis of Gravity and Magnetic Anomalies

Identifying presence of correlated errors using machine learning algorithms for the selective de-correlation of GRACE harmonic coefficients

The Electric Network Frequency (ENF) criterion provides useful forensic evidence for multimedia authentication. In this paper, a systematic study of non-parametric and parametric spectral estimation methods is conducted for ENF extraction. Fast implementations of the Capon method and the Iterative Adaptive Approach, which exploit the Gohberg-Semencul factorization of the inverse covariance matrix, are included as well. When long segments are used, a very high matching accuracy is achieved. That is, the maximum correlation-coefficient between the extracted ENF and the ground truth may exceed 99%. Similarly, the standard deviation of error maybe as small as 1.069 · 10-3. Non-parametric spectral estimation techniques are shown to be able to detect an alteration in an audio recording, where a short utterance recorded in Europe is replaced by the same content recorded in the US.

Assessing spectral estimation methods for electric network frequency extraction

Blackman–Tukey spectral estimation and electric network frequency matching from power mains and speech recordings

An efficient learning scheme is proposed for hypergraph ranking based on multiple optimizations. This scheme dynamically optimizes the hypergraph structure through the incidence matrix and employs adaptive hyperedge weight estimation based on gradient descent method. By doing so, optimized hypergraph ranking vectors are derived, which recommend images of places of interest (POIs) worth visiting or tags to images associated to POIs. Experiments are conducted on a dataset of Greek POIs crawled from F l i c k r and the NUS-WIDE-LITE dataset. The proposed method offers higher quality recommendations than state-of-the-art methods, employing either a static hypergraph structure, a hypergraph structure learning, or a hyperedge weight update with respect to precision.

Adaptive hypergraph learning with multi-stage optimizations for image and tag recommendation

Electric Network Frequency (ENF) is embedded in multimedia recordings if the recordings are captured with a device connected to power mains or placed near the power mains. It is exploited as a tool for multimedia authentication. ENF fluctuates stochastically around its nominal frequency at 50/60 Hz. In indoor environments, luminance variations captured by video recordings can also be exploited for ENF estimation. However, the various textures and different levels of shadow and luminance hinder ENF estimation in static and non-static video, making it a non-trivial problem. To address this problem, a novel automated approach is proposed for ENF estimation in static and non-static digital video recordings. The proposed approach is based on the exploitation of areas with similar characteristics in each video frame. These areas, called superpixels, have a mean intensity that exceeds a specific threshold. The performance of the proposed approach is tested on various videos of real-life scenarios that resemble surveillance from security cameras. These videos are of escalating difficulty and span recordings from static ones to recordings, which exhibit continuous motion. The maximum correlation coefficient is employed to measure the accuracy of ENF estimation against the ground truth signal. Experimental results show that the proposed approach improves ENF estimation against the state-of-the-art, yielding statistically significant accuracy improvements.

Georgios Karantaidis

Papers

Identifying presence of correlated errors using machine learning algorithms for the selective de-correlation of GRACE harmonic coefficients

Assessing spectral estimation methods for electric network frequency extraction

Blackman–Tukey spectral estimation and electric network frequency matching from power mains and speech recordings

Adaptive hypergraph learning with multi-stage optimizations for image and tag recommendation

An Automated Approach for Electric Network Frequency Estimation in Static and Non-Static Digital Video Recordings.