Preprocessing Layer Research Articles

Deep multilabel CNN for forensic footwear impression descriptor identification

In recent years deep neural networks have become the workhorse of computer vision. In this paper, we employ a deep learning approach to classify footwear impression’s features known as descriptors for forensic use cases. Within this process, we develop and evaluate an effective technique for feeding downsampled greyscale impressions to a neural network pre-trained on data from a different domain. Our approach relies on learnable preprocessing layer paired with multiple interpolation methods used in parallel. We empirically show that this technique outperforms using a single type of interpolated image without learnable preprocessing, and can help to avoid the computational penalty related to using high resolution inputs, by making more efficient use of the low resolution inputs. We also investigate the effect of preserving the aspect ratio of the inputs, which leads to considerable boost in accuracy without increasing the computational budget with respect to squished rectangular images. Finally, we formulate a set of best practices for transfer learning with greyscale inputs, potentially widely applicable in computer vision tasks ranging from footwear impression classification to medical imaging.

General Purpose Image Tampering Detection using Convolutional Neural Network and Local Optimal Oriented Pattern (LOOP)

Digital image tampering detection has been an active area of research in recent times due to the ease with which digital image can be modified to convey false or misleading information. To address this problem, several studies have proposed forensics algorithms for digital image tampering detection. While these approaches have shown remarkable improvement, most of them only focused on detecting a specific type of image tampering. The limitation of these approaches is that new forensic method must be designed for each new manipulation approach that is developed. Consequently, there is a need to develop methods capable of detecting multiple tampering operations. In this paper, we proposed a novel general purpose image tampering scheme based on CNNs and Local Optimal Oriented Pattern (LOOP) which is capable of detecting five types of image tampering in both binary and multiclass scenarios. Unlike the existing deep learning techniques which used constrained pre-processing layers to suppress the effect of image content in order to capture image tampering traces, our method uses LOOP features, which can effectively subdue the effect image content, thus, allowing the proposed CNNs to capture the needed features to distinguish among different types of image tampering. Through a number of detailed experiments, our results demonstrate that the proposed general purpose image tampering method can achieve high detection accuracies in individual and multiclass image tampering detections respectively and a comparative analysis of our results with the existing state of the arts reveals that the proposed model is more robust than most of the exiting methods.

Blind detection of glow-based facial forgery

With the rapid development of artificial intelligence technologies, various generative models can synthesize fake face images with photo-realistic effects. Glow, a generative flow using invertible 1×1 convolution, is a state-of-the-art technique for efficient synthesis of face images with high resolution and fidelity. However, facial forgeries bring serious challenges to morality, ethics and public confidence. Especially, facial forgeries might change the semantic content conveyed by a face image. A Convolutional Neural Network (CNN) based model, namely SCnet, is proposed to expose the Glow-based facial forgery. Specifically, an image sharpening operator is embedded in the convolutional layer as the pre-processing layer of the network to highlight the traces left by Glow. Then, SCnet is specifically designed to automatically learn high-level forensics features from pre-processing results. Moreover, a fake face dataset is built by exploiting the CelebA face image dataset and the Glow-based forgery technique. A series of experiments are conducted to prove the effectiveness of the proposed approach. Experimental results show that the proposed approach achieves a classification accuracy up to 95.92% under various post-processing operations.

JPEG steganalysis based on ResNeXt with Gauss partial derivative filters

The latest research indicates that the image steganalysis has been greatly promoted by convolutional neural networks (CNNs). This study further addresses the problem of JPEG steganalysis through proposing a novel CNN architecture in which Gauss partial derivative (GPD) filters and two constructed blocks based on ResNeXt are integrated. In the proposed network, multi-order GPD filters are designed as the pre-processing layer to generate residual images, which can effectively capture sufficient embedding disturbance in texture and edge regions. Furthermore, referring to ResNeXt, two multi-branch blocks are constructed and aggregated to fully exploit the residual images to generate image features for classification. Numerous experiments have been conducted against J-UNIWARD on the public dataset to demonstrate the effectiveness and remarkable performance of the proposed network. Experimental results prove that the proposed network makes better performance than state-of-the-art CNN-based method J-Xu-Net and SCA-GFR. Source code is available via GitHub: https://github.com/Ante-Su/RXGNet .

Image Edge Detector with Gabor Type Filters Using a Spiking Neural Network of Biologically Inspired Neurons

We report the design of a Spiking Neural Network (SNN) edge detector with biologically inspired neurons that has a conceptual similarity with both Hodgkin-Huxley (HH) model neurons and Leaky Integrate-and-Fire (LIF) neurons. The computation of the membrane potential, which is used to determine the occurrence or absence of spike events, at each time step, is carried out by using the analytical solution to a simplified version of the HH neuron model. We find that the SNN based edge detector detects more edge pixels in images than those obtained by a Sobel edge detector. We designed a pipeline for image classification with a low-exposure frame simulation layer, SNN edge detection layers as pre-processing layers and a Convolutional Neural Network (CNN) as a classification module. We tested this pipeline for the task of classification with the Digits dataset, which is available in MATLAB. We find that the SNN based edge detection layer increases the image classification accuracy at lower exposure times, that is, for 1 < t < T /4, where t is the number of milliseconds in a simulated exposure frame and T is the total exposure time, with reference to a Sobel edge or Canny edge detection layer in the pipeline. These results pave the way for developing novel cognitive neuromorphic computing architectures for millisecond timescale detection and object classification applications using event or spike cameras.

Rule-Based Actionable Intelligence for Disaster Situation Management

Managing natural disasters is a social responsibility as they might cause a gloomy impact on human life. Efficient and timely alert systems for public and actionable recommendations for decision makers may well decrease the number of casualties. Web semantics strengthen the description of web resources for exploiting them better and making them more meaningful for both human and machine. In this work, the authors propose a semantic rule-based approach for disaster situation management (DSM) to reach the next level of decision-making power and its architecture for providing actionable intelligence in the domain of the earthquake. The system itself is based on a data pre-processing layer, a computation layer, and the middle layer relies on an extensive rule base of experts' advice stored over time and a disaster ontology along with its inherent semantics. The rule-based reasoning approach uses this knowledge base in combination with the expert rule base, written in SWRL rules, to infer recommendations for the response to an earthquake.

Research on Target-Driven Navigation of Mobile Robot Based on Deep Reinforcement Learning and Preprocessing Layer

Recently, with the rise of deep reinforcement learning model, robot navigation based on this method has a huge advantage compared with traditional slam method, which has attracted extensive attention. However, when the navigation algorithm trained in the virtual environment is transferred to the real environment, the navigation performance of the robot will decline sharply because of the great difference between the virtual environment and the real environment. In order to improve the navigation ability of mobile robot, this paper implements a mobile robot navigation system based on deep reinforcement learning without environment map and only visual input. At the same time, in order to solve the problem of poor generalization ability of deep reinforcement learning from virtual environment to real environment, this paper proposes a preprocessing layer with knowledge and combines it with deep reinforcement learning module. The combined algorithm model alleviates the performance fault problem caused by the migration algorithm and the performance difference between virtual sensor and real sensor. At the end of this paper, a navigation experiment based on the turtlebot is designed, which proves that the deep reinforcement learning algorithm with the preprocessing layer can alleviate the performance fault problem caused by the migration algorithm, and have a certain ability of obstacle avoidance and avoidance without the environment map.

Task-Oriented Intelligent Networking Architecture for the Space–Air–Ground–Aqua Integrated Network

As one of the most promising networks, the space–air–ground–aqua integrated network (SAGAIN) has the characteristics of wide coverage and large information capacity, which can meet various requests from users in different domains. With the rapid growth of data and information generated by the Internet of Things (IoT), SAGAIN has received much attention in recent years. However, the existing network architectures are not capable of providing personalized network services according to different task types in SAGAIN. Besides, they cannot deal with many problems in SAGAIN well, such as heterogeneous network disconnection, high network delay, intermittent interruption, and unbalanced network load. In this article, in order to solve the abovementioned problems, we propose a novel architecture for SAGAIN named task-oriented intelligent networking architecture (TOINA). First, we apply the edge-cloud computing technology and network domain division in TOINA to realize intelligent networking and reduce the latency. Second, the task-oriented networking method is proposed to provide personalized network services and increase network intelligence. Third, we intend to leverage the information center network (ICN) paradigm to build the SAGAIN and optimize the content naming rules. Furthermore, a preprocessing layer was added in the network protocol stack to perform the heterogeneous network convergence in SAGAIN. In addition, some security technologies related to network architecture are considered in SAGAIN. This article presents the background, rationale, and benefits of the TOINA for SAGAIN. Besides, a specific case is studied to illustrate the network architecture work process further.

Hierarchical growing grid networks for skeleton based action recognition

In this paper, a novel cognitive architecture for action recognition is developed by applying layers of growing grid neural networks. Using these layers makes the system capable of automatically arranging its representational structure. In addition to the expansion of the neural map during the growth phase, the system is provided with a prior knowledge of the input space, which increases the processing speed of the learning phase. Apart from two layers of growing grid networks the architecture is composed of a preprocessing layer, an ordered vector representation layer and a one-layer supervised neural network. These layers are designed to solve the action recognition problem. The first-layer growing grid receives the input data of human actions and the neural map generates an action pattern vector representing each action sequence by connecting the elicited activation of the trained map. The pattern vectors are then sent to the ordered vector representation layer to build the time-invariant input vectors of key activations for the second-layer growing grid. The second-layer growing grid categorizes the input vectors to the corresponding action clusters/sub-clusters and finally the one-layer supervised neural network labels the shaped clusters with action labels. Three experiments using different datasets of actions show that the system is capable of learning to categorize the actions quickly and efficiently. The performance of the growing grid architecture is compared with the results from a system based on Self-Organizing Maps, showing that the growing grid architecture performs significantly superior on the action recognition tasks.

A Steganalysis framework based on CNN using the filter subset selection method

The steganalysis method based on Convolutional Neural Network (CNN) has attracted wide attention in the field of steganalysis. This method typically uses high-pass or derivative filters to pre-process images. Multiple filters can be used to improve the detection accuracy. However, the use of multiple filters may generate redundant residual images and redundant features. Redundant features not only consume computing resources and time but also cause model over-fitting, thus compromising the detection accuracy of the model. Therefore, we proposed a filter subset selection method to develop a well-designed pre-processing layer for CNN-based steganalysis framework. This method discarded many high-pass and derivative filters according to the mechanism of convolution operation and the correlations between pixels. Structural Similarity (SSIM) was used to calculate the similarity between the filtered residual images and arrange them in ascending order. Finally, based on the arranged filters, a series of experiments were conducted to determine the optimal filter subset and the optimal CNN-based steganalysis framework. The experimental results indicate that the proposed method not only guarantees detection accuracy but also improves the training efficiency of the model. Therefore, this method offers an optimized trade-off between computational complexity and detection accuracy.

Enhancing deep discriminative feature maps via perturbation for face presentation attack detection

Face presentation attack detection (PAD) in unconstrained conditions is one of the key issues in face biometric-based authentication and security applications. In this paper, we propose a perturbation layer — a learnable pre-processing layer for low-level deep features — to enhance the discriminative ability of deep features in face PAD. The perturbation layer takes the deep features of a candidate layer in Convolutional Neural Network (CNN), the corresponding hand-crafted features of an input image, and produces adaptive convolutional weights for the deep features of the candidate layer. These adaptive convolutional weights determine the importance of the pixels in the deep features of the candidate layer for face PAD. The proposed perturbation layer adds very little overhead to the total trainable parameters in the model. We evaluated the proposed perturbation layer with Local Binary Patterns (LBP), with and without color information, on three publicly available face PAD databases, i.e., CASIA, Idiap Replay-Attack, and OULU-NPU databases. Our experimental results show that the introduction of the proposed perturbation layer in the CNN improved the face PAD performance, in both intra-database and cross-database scenarios. Our results also highlight the attention created by the proposed perturbation layer in the deep features and its effectiveness for face PAD in general.

Deep convolutional neural network-based feature extraction for steganalysis of content-adaptive JPEG steganography

Deep learning-based steganalysis techniques have significantly progressed in recent years. Compared with typical steganalysis methods using rich model features and classifiers, deep learning-based steganalysis methods usually train a deep convolutional neural network (CNN) as the steganography detector. One advantage of a deep CNN is the strong capability for discriminative feature learning. The input feature map of the fully connected layer is the steganalysis feature learned using the deep CNN. We can, therefore, extract the steganalysis feature based on the trained deep CNN. Compared with steganalysis features constructed by hand, the extraction of steganalysis features using a deep CNN can take advantage of the strong feature-learning capability of such a network. Therefore, two types of typical steganalysis frameworks are first compared. Then, a deep CNN for steganalysis feature learning is constructed, and the setting of the image preprocessing layer is discussed. Next, the detection performances of the different learned steganalysis features are compared. Finally, the detailed extraction process of the proposed learned steganalysis feature is described. The experiment results show that the proposed steganalysis method, combining learned and handcrafted steganalysis features, can significantly improve the detection performance for content-adaptive JPEG steganography.

Commercial Devices-Based System Designed to Improve the Treatment Adherence of Hypertensive Patients.

This paper presents an intelligent system designed to increase the treatment adherence of hypertensive patients. The architecture was developed to allow communication among patients, physicians, and families to determine each patient’s medication intake and self-monitoring of blood pressure rates. Concerning the medication schedule, the system is designed to follow a predefined prescription, adapting itself to undesired events, such as mistakenly taking medication or forgetting to take medication on time. When covering the blood pressure measurement, it incorporates best medical practices, registering the actual values in recommended frequency and form, trying to avoid the known “white-coat effect.” We assume that taking medicine precisely and measuring blood pressure correctly may lead to good adherence to the treatment. The system uses commercial consumer electronic devices and can be replicated in any home equipped with a standard personal computer and Internet access. The resulting architecture has four layers. The first is responsible for adding electronic devices that typically exist in today’s homes to the system. The second is a preprocessing layer that filters the data generated from the patient’s behavior. The third is a reasoning layer that decides how to act based on the patient’s activities observed. Finally, the fourth layer creates messages that should drive the reactions of all involved actors. The reasoning layer takes into consideration the patient’s schedule and medication-taking activity data and uses implicit algorithms based on the J48, RepTree, and RandomTree decision tree models to infer the adherence. The algorithms were first adjusted using one academic machine learning and data mining tool. The system communicates with users through smartphones (anytime and anywhere) and smart TVs (in the patient’s home) by using the 3G/4G and WiFi infrastructure. It interacts automatically through social networks with doctors and relatives when changes or mistakes in medication intake and blood pressure mean values are detected. By associating the blood pressure data with the history of medication intake, our system can indicate the treatment adherence and help patients to achieve better treatment results. Comparisons with similar research were made, highlighting our findings.

Convolutional Neural Network for Copy-Move Forgery Detection

Digital image forgery is a growing problem due to the increase in readily-available technology that makes the process relatively easy. In response, several approaches have been developed for detecting digital forgeries. This paper proposes a novel scheme based on neural networks and deep learning, focusing on the convolutional neural network (CNN) architecture approach to enhance a copy-move forgery detection. The proposed approach employs a CNN architecture that incorporates pre-processing layers to give satisfactory results. In addition, the possibility of using this model for various copy-move forgery techniques is explained. The experiments show that the overall validation accuracy is 90%, with a set iteration limit.

A Data Driven Multi-Layer Framework of Pervasive Information Computing System for eHealthcare

In the last decade, significant advancements in telecommunications and informatics have seen which incredibly boost mobile communications, wireless networks, and pervasive computing. It enables healthcare applications to increase human livelihood. Furthermore, it seems feasible to continuous observation of patients and elderly individuals for their wellbeing. Such pervasive arrangements enable medical experts to analyse current patient status, minimise reaction time, increase livelihood, scalability, and availability. There is found plenty of remote patient monitoring model in literature, and most of them are designed with limited scope. Most of them are lacking to give an overall unified, complete model which talk about all state-of-the-art functionalities. In this regard, remote patient monitoring systems (RPMS's) play important roles through wearable devices to monitor the patient's physiological condition. RPMS also enables the capture of related videos, images, and frames. RPMS do not mean to enable only capturing various sorts of patient-related information, but it also must facilitate analytics, transformation, security, alerts, accessibility, etc. In this view, RPMS must ensure some broad issues like, wearability, adaptability, interoperability, integration, security, and network efficiency. This article proposes a data-driven multi-layer architecture for pervasively remote patient monitoring that incorporates these issues. The system has been classified into five fundamental layers: the data acquisition layer, the data pre-processing layer, the network and data transfer layer, the data management layer and the data accessing layer. It enables patient care at real-time using the network infrastructure efficiently. A detailed discussion on various security issues have been carried out. Moreover, standard deviation-based data reduction and a machine-learning-based data access policy is also proposed.

Depth-Wise Separable Convolutions and Multi-Level Pooling for an Efficient Spatial CNN-Based Steganalysis

For steganalysis, many studies showed that convolutional neural network (CNN) has better performances than the two-part structure of traditional machine learning methods. Existing CNN architectures use various tricks to improve the performance of steganalysis, such as fixed convolutional kernels, the absolute value layer, data augmentation and the domain knowledge. However, some designing of the network structure were not extensively studied so far, such as different convolutions (inception, xception, etc.) and variety ways of pooling(spatial pyramid pooling, etc.). In this paper, we focus on designing a new CNN network structure to improve detection accuracy of spatial-domain steganography. First, we use $3\times 3$ kernels instead of the traditional $5\times 5$ kernels and optimize convolution kernels in the preprocessing layer. The smaller convolution kernels are used to reduce the number of parameters and model the features in a small local region. Next, we use separable convolutions to utilize channel correlation of the residuals, compress the image content and increase the signal-to-noise ratio (between the stego signal and the image signal). Then, we use spatial pyramid pooling (SPP) to aggregate the local features and enhance the representation ability of features by multi-level pooling. Finally, data augmentation is adopted to further improve network performance. The experimental results show that the proposed CNN structure is significantly better than other five methods such as SRM, Ye-Net, Xu-Net, Yedroudj-Net and SRNet, when it is used to detect three spatial algorithms such as WOW, S-UNIWARD and HILL with a wide variety of datasets and payloads.

736 Published By: Blue Eyes Intelligence Engineering & Sciences Publication Retrieval Number: B1780078219/19©BEIESP DOI: 10.35940/ijrte.B1780.078219 Journal Website: www.ijrte.org A Meta Classification Model for Stegoanalysis using Generic NN

The core idea behind deep learning is that comprehensive feature representations can be efficiently learned with the deep architectures which are collected of stacked layer of trainable non linear operation. However, because of the diversity of image content, it is hard to learn effective feature representations directly from images for steGAnalysis. SteGAnalysis may be generally figured as binary classification issue. This technique, which is called a universal/blind steGAnalysis, will become the principle stream around current steGAnalytic algorithms. In the preparation phase, effective features which are sensitive with message embedding are concentrated on highlight possibility control by steGAnographier. Then, a binary classifier will be discovered looking into pairs from claiming blanket pictures and their relating stegos pointing with Figure a limit on recognize steGAnography. On testing phase, those prepared classifier is used to anticipate labels from claiming new enter pictures. Past exploration indicated that it will be rather critical to power spread Characteristics Also stego offers to be paired, i. e. SteGAnalytic offers from claiming spread pictures And their stego pictures ought further bolstering be safeguarded in the preparing situated. Otherwise, breaking cover-stego pairs in distinctive sets might present biased error and prompt to a suboptimal execution. Proposed approaches have to fix the kernel of first layer as the HPF (high-pass filter). It is so-called pre-processing layer. We suggested another technic with characteristic decrease done which characteristic Choice and extraction And classifier preparation need aid performed at the same time utilizing a generic calculation. That generic calculation optimizes An characteristic weight vector used to scale the individual features in the unique example vectors. A masker vector may be likewise utilized to concurrent Choice of a characteristic subset. We utilize this technobabble clinched alongside mix with those RESNET, and look at the outcomes with established characteristic Choice and extraction systems.

RestegNet: a residual steganalytic network

Deep convolutional networks bring new energy to image steganography. It is an opportunity for steganalysis research. However, the operations to widen the gap between covers and stegos are only in the preprocessing layers for most existing networks. In this paper, a residual steganalytic network (RestegNet) is proposed to overcome this limitation. We design a novel building block group, which consists of two alternating building blocks: 1) A sharpening block based on residual connections (ShRC), which makes the noise of steganography overwhelm the image content, and aims to enhance steganographic signal detectability. 2) A smoothing block based on residual connections (SmRC), which seeks to downsample the feature maps to boil them down to useful data. First, we use the same preprocessing layers as previous methods to ensure minimum performance. Then, we use these building block groups to exaggerate the traces of steganography further and make the difference between covers and stegos in the feature extraction layers. Contrastive experiments with previous methods conducted on the BOSSbase 1.01 demonstrate the effectiveness and the superior performance of the proposed network.

A Framework for Prediction of Household Energy Consumption Using Feed Forward Back Propagation Neural Network

Energy is considered the most costly and scarce resource, and demand for it is increasing daily. Globally, a significant amount of energy is consumed in residential buildings, i.e., 30–40% of total energy consumption. An active energy prediction system is highly desirable for efficient energy production and utilization. In this paper, we have proposed a methodology to predict short-term energy consumption in a residential building. The proposed methodology consisted of four different layers, namely data acquisition, preprocessing, prediction, and performance evaluation. For experimental analysis, real data collected from 4 multi-storied buildings situated in Seoul, South Korea, has been used. The collected data is provided as input to the data acquisition layer. In the pre-processing layer afterwards, several data cleaning and preprocessing schemes are applied to the input data for the removal of abnormalities. Preprocessing further consisted of two processes, namely the computation of statistical moments (mean, variance, skewness, and kurtosis) and data normalization. In the prediction layer, the feed forward back propagation neural network has been used on normalized data and data with statistical moments. In the performance evaluation layer, the mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean squared error (RMSE) have been used to measure the performance of the proposed approach. The average values for data with statistical moments of MAE, MAPE, and RMSE are 4.3266, 11.9617, and 5.4625 respectively. These values of the statistical measures for data with statistical moments are less as compared to simple data and normalized data which indicates that the performance of the feed forward back propagation neural network (FFBPNN) on data with statistical moments is better when compared to simple data and normalized data.

Contrasting Advantages of Learning With Random Weights and Backpropagation in Non-Volatile Memory Neural Networks

Recently, a Cambrian explosion of a novel, non-volatile memory (NVM) devices known as memristive devices have inspired effort in building hardware neural networks that learn like the brain. Early experimental prototypes built simple perceptrons from nanosynapses, and recently, fully-connected multi-layer perceptron (MLP) learning systems have been realized. However, while backpropagating learning systems pair well with high-precision computer memories and achieve state-of-the-art performances, this typically comes with a massive energy budget. For future Internet of Things/peripheral use cases, system energy footprint will be a major constraint, and emerging NVM devices may fill the gap by sacrificing high bit precision for lower energy. In this paper, we contrast the well-known MLP approach with the extreme learning machine (ELM) or NoProp approach, which uses a large layer of random weights to improve the separability of high-dimensional tasks, and is usually considered inferior in a software context. However, we find that when taking the device non-linearity into account, NoProp manages to equal hardware MLP system in terms of accuracy. While also using a sign-based adaptation of the delta rule for energy-savings, we find that NoProp can learn effectively with four to six ’bits’ of device analog capacity, while MLP requires eight-bit capacity with the same rule. This may allow the requirements for memristive devices to be relaxed in the context of online learning. By comparing the energy footprint of these systems for several candidate nanosynapses and realistic peripherals, we confirm that memristive NoProp systems save energy compared with MLP systems. Lastly, we show that ELM/NoProp systems can achieve better generalization abilities than nanosynaptic MLP systems when paired with pre-processing layers (which do not require backpropagated error). Collectively, these advantages make such systems worthy of consideration in future accelerators or embedded hardware.