Discriminative Feature Vectors Research Articles

Video Forgery Detection using an Improved BAT with Stacked Auto Encoder Model

There are various public and private places such as banks, roads, offices, and homes equipped with cameras for surveillance. The surveillance videos are consisting of a precious source of information related to critical application scopes. The main problem is to aid powerful and accessible software that changes the content present in the video for the forgery creation of a video. The forgery involves region duplication that has a common video tampering. The existing techniques are utilized to detect video tampering from the forged videos that showed complexity in the background. Thus, it is important to overcome the problem of forgery detection in the research. The Spatio-temporal averaging model is carried out for the collection of a video sequence for obtaining the background information. This can detect the moving objects effectively for forgery detection. Next, the ResNet 18 is used for extraction of the feature vectors, and the discriminative feature vectors were reduced and improved the training time and accuracy. The Single Auto Encoder (SAE) is not able to reduce the input features' dimensionality. Thus, the SAE has used 3 encoders stacked on the top for detecting the forgery. It is based on the sequence of videos. In comparison to the existing models, the proposed approach outperformed them with accuracy rates of 98.6%, sensitivity rates of 98.60%, specificity rates of 98.47%, MCC rates of 97.29%, and precision rates of 99.93%.

Melanoma skin cancer detection using mask-RCNN with modified GRU model.

Introduction: Melanoma Skin Cancer (MSC) is a type of cancer in the human body; therefore, early disease diagnosis is essential for reducing the mortality rate. However, dermoscopic image analysis poses challenges due to factors such as color illumination, light reflections, and the varying sizes and shapes of lesions. To overcome these challenges, an automated framework is proposed in this manuscript. Methods: Initially, dermoscopic images are acquired from two online benchmark datasets: International Skin Imaging Collaboration (ISIC) 2020 and Human against Machine (HAM) 10000. Subsequently, a normalization technique is employed on the dermoscopic images to decrease noise impact, outliers, and variations in the pixels. Furthermore, cancerous regions in the pre-processed images are segmented utilizing the mask-faster Region based Convolutional Neural Network (RCNN) model. The mask-RCNN model offers precise pixellevel segmentation by accurately delineating object boundaries. From the partitioned cancerous regions, discriminative feature vectors are extracted by applying three pre-trained CNN models, namely ResNeXt101, Xception, and InceptionV3. These feature vectors are passed into the modified Gated Recurrent Unit (GRU) model for MSC classification. In the modified GRU model, a swish-Rectified Linear Unit (ReLU) activation function is incorporated that efficiently stabilizes the learning process with better convergence rate during training. Results and discussion: The empirical investigation demonstrate that the modified GRU model attained an accuracy of 99.95% and 99.98% on the ISIC 2020 and HAM 10000 datasets, where the obtained results surpass the conventional detection models.

Circular shift combination local binary pattern (CSC-LBP) method for dorsal finger crease classification

Biometric technology has drawn increasing attention and significance importance in recent years. In biometric security systems, personal identification and verification rely on their physical, behavioral, and biological characteristics. In this study, a new hand-based modality called dorsal finger creases is proposed for biometric classification. This modality is located on the dorsal surface of the finger, between the proximal knuckle and distal knuckle of the finger. However, it requires a specific feature extraction approach to extract the modality information on the selected region. Therefore, we have proposed a method for extracting the underlying features of the dorsal finger creases, called circular shift combination local binary pattern (CSC-LBP). The concept of CSC-LBP is to compute the local binary pattern within a 3 × 3 spatial window for each neighborhood pixel separately. Further, the concept of combination approach is applied on the individually computed eight LBP values to obtain the more discriminative feature vector. A multiclass support vector machine classifier is used for evaluating the effectiveness of the proposed CSC-LBP operator. Extensive experiments on self-collected datasets demonstrate the high classification accuracy and effectiveness of the proposed CSC-LBP method and confirm the usefulness of dorsal finger creases for personal recognition.

Efficient private key generation from iris data for privacy and security applications

Private key generation based on biometric data is gaining popularity in several cryptographic applications. Some key generation approaches using fingerprint and face data are known for this purpose. The existing approaches, in general, follow direct features from raw data, which may have an issue of revealing users’ biometric data. Further, iris biometrics being a better source for stable and secure key generation, is yet to be explored. In this work, an iris biometric-based key generation approach is proposed. In the proposed method, first, an ensemble L0 Gradient Minimization and an edge-preserving filter are used to extract iris structure from the iris images. Secondly, the iris texture data is normalized, and the consistent region is selected based on a novel statistical method. Thirdly, feature vectors are generated using the ensemble local space-filling curve descriptors and their variants. Fourth, a discriminant feature vector is generated using hybrid feature selection and neighbourhood component analysis. Finally, an interval-based encoding scheme is used to generate a key from the discriminant feature vector. To substantiate the efficacy of the proposed approach, extensive experiments have been carried out with near-infrared images, visible wavelength images, at-a-distance, and on-move images. Further, distinctiveness, dissimilarity, randomness, and security analyses have been carried out to validate the resilience of keys against different attacks. More significantly, a unique key can be generated dynamically, that is, from an online image. The proposed approach is applicable to different cryptographic applications, such as data storage security, remote authentication protocol, blockchain system security, etc.

Automated Urine Cell Image Classification Model Using Chaotic Mixer Deep Feature Extraction.

Microscopic examination of urinary sediments is a common laboratory procedure. Automated image-based classification of urinary sediments can reduce analysis time and costs. Inspired by cryptographic mixing protocols and computer vision, we developed an image classification model that combines a novel Arnold Cat Map (ACM)- and fixed-size patch-based mixer algorithm with transfer learning for deep feature extraction. Our study dataset comprised 6,687 urinary sediment images belonging to seven classes: Cast, Crystal, Epithelia, Epithelial nuclei, Erythrocyte, Leukocyte, and Mycete. The developed model consists of four layers: (1) an ACM-based mixer to generate mixed images from resized 224 ×224 input images using fixed-size 16 ×16 patches; (2) DenseNet201 pre-trained on ImageNet1K to extract 1,920 features from each raw input image, and its six corresponding mixed images were concatenated to form a final feature vector of length 13,440; (3) iterative neighborhood component analysis to select the most discriminative feature vector of optimal length 342, determined using a k-nearest neighbor (kNN)-based loss function calculator; and (4) shallow kNN-based classification with ten-fold cross-validation. Our model achieved 98.52% overall accuracy for seven-class classification, outperforming published models for urinary cell and sediment analysis. We demonstrated the feasibility and accuracy of deep feature engineering using an ACM-based mixer algorithm for image preprocessing combined with pre-trained DenseNet201 for feature extraction. The classification model was both demonstrably accurate and computationally lightweight, making it ready for implementation in real-world image-based urine sediment analysis applications.

Feature Fusion and NRML Metric Learning for Facial Kinship Verification

Features extracted from facial images are used in various fields such as kinship verification. The kinship verification system determines the kin or non-kin relation between a pair of facial images by analysing their facial features. In this research, different texture and color features have been used along with the metric learning method, to verify the kinship for the four kinship relations of father-son, father-daughter, mother-son and mother-daughter. First, by fusing effective features, NRML metric learning used to generate the discriminative feature vector, then SVM classifier used to verify to kinship relations. To measure the accuracy of the proposed method, KinFaceW-I and KinFaceW-II databases have been used. The results of the evaluations show that the feature fusion and NRML metric learning methods have been able to improve the performance of the kinship verification system. In addition to the proposed approach, the effect of feature extraction from the image blocks or the whole image is investigated and the results are presented. The results indicate that feature extraction in block form, can be effective in improving the final accuracy of kinship verification.

Facial expression recognition based on regional adaptive correlation

AbstractTo address the problem that the features extracted by CNN‐based facial expression recognition (FER) do not consider structural information, a region adaptive correlation deep network (RACN) is proposed. The network consists of two branches. In one branch, the features obtained by applying CNN to facial sub‐blocks are used as the input of the proposed second‐order region correlation network (SRCN), which obtains structural features by adaptively learning the correlation of facial regions. Furthermore, they are fused with the parallel branch‐extracted global features to obtain a comprehensive high‐semantic feature representation. Finally, weights are assigned to the two features through the channel attention mechanism for more accurate expression classification. Experimental results show that our method can effectively extract expression features in an end‐to‐end manner, improve the accuracy of FER, and achieve competitive performance without relying on any a priori knowledge. And the region‐adaptive correlation feature extraction branch RACN can be applied to other deep learning networks to extract discriminative structural‐adaptive features. To the best of our knowledge, our work is the first to enrich the feature representation for end‐to‐end static FER by adaptively obtaining more discriminative regional adaptive correlation feature vectors via the autocorrelation matrix combined with CNN compared to the existing literature.

SMAM: Self and Mutual Adaptive Matching for Skeleton-Based Few-Shot Action Recognition.

This paper focuses on skeleton-based few-shot action recognition. Since skeleton is essentially a sparse representation of human action, the feature maps extracted from it, through a standard encoder network in the few-shot condition, may not be sufficiently discriminative for some action sequences that look partially similar to each other. To address this issue, we propose a self and mutual adaptive matching (SMAM) module to convert such feature maps into more discriminative feature vectors. Our method, named as SMAM-Net, first leverages both the temporal information associated with each individual skeleton joint and the spatial relationship among them for feature extraction. Then, the SMAM module adaptively measures the similarity between labeled and query samples and further carries out feature matching within the query set to distinguish similar skeletons of various action categories. Experimental results show that the SMAM-Net outperforms other baselines on the large-scale NTU RGB+D 120 dataset in the tasks of one-shot and five-shot action recognition. We also report our results on smaller datasets including NTU RGB+D 60, SYSU and PKU-MMD to demonstrate that our method is reliable and generalises well on different datasets. Codes and the pretrained SMAM-Net will be made publicly available.

Enhanced grasshopper optimization algorithm with extreme learning machines for motor‐imagery classification

AbstractIn Brain Computer Interface (BCI), achieving a reliable motor‐imagery classification is a challenging task. The set of discriminative and relevant feature vectors plays a crucial role in classification. In this article, an enhanced optimization technique is implemented for selecting active feature vectors to enhance motor‐imagery classification using Electroencephalography (EEG) signals. After collecting the input EEG signals from BCI competition III‐4a and IV‐2a databases, the 6th‐order butter‐worth filter is employed for eliminating base‐line wander noise from the raw EEG signals. Further, the Variational Mode Decomposition technique is applied for separating the important signal components from the composite EEG signals, and then, the Higher Order Statistic, kurtosis, skewness, standard deviation, and entropy are utilized for feature extraction. The high‐dimensional feature values are given to the Enhanced Grasshopper Optimization Algorithm for optimum feature selection, which are given to the Extreme Learning Machines (ELM) classifier for motor‐imagery classification. Finally, in the resulting section, the optimized ELM model achieved 99.48% and 99.12% of accuracy on the BCI competition III‐4a and IV‐2a databases, where the achieved results are maximum compared to the traditional deep learning models.

ECG beat classification based on discriminative multilevel feature analysis and deep learning approach

Extraction of significant features from Electrocardiogram (ECG) signal is the primary concern for accurate diagnosis of cardiac arrhythmia. This work presents a novel approach of multilevel feature analysis and deep learning strategy for efficient ECG beat classification. The multilayer characteristics of ECG signal obtained from Empirical mode decomposition (EMD) are explored to extract discriminative feature vectors. The multilayer similarity coefficients are obtained by applying Dynamic time warping (DTW) metric and Pearson correlation coefficient (PCC) as diagnostic features. Furthermore, discrete orthonormal Stockwell transform (DOST) is employed for time–frequency representation of ECG data in multilayer aspect. The sublet changes in time–frequency spaces due to the presence of cardiac abnormalities are captured by estimating various nonlinear parameters. Interlayer deviations of these nonlinear parameters are estimated as the significant characteristics of arrhythmia detection. In addition, this study shows that the phase synchrony (PS) coefficients are prominent index for quantifying the crucial phase variation between normal and abnormal heart conditions. Hence multilevel PS coefficients are employed as the predictors of arrhythmia detection. Finally, the extracted feature vectors are fed to various classifiers to identify the heart anomalies. The proposed technique attains average accuracy of 98.82% and 98.14% using support vector machine (SVM) and k-nearest neighbors (k-NN) classifier respectively. The improved classification accuracy of 99.05% is obtained with the strategy of combining deep neural network (DNN) with the proposed feature extraction policy. Present work delivers satisfactory and superior performances for arrhythmia classification compare to other existing approaches.

Identifying defective solar cells in electroluminescence images using deep feature representations.

Electroluminescence (EL) imaging is a technique for acquiring images of photovoltaic (PV) modules and examining them for surface defects. Analysis of EL images has been manually performed by visual inspection of images by experts. This manual procedure is tedious, time-consuming, subjective, and requires deep expert knowledge. In this work, a hybrid and fully-automated classification system is developed for detecting different types of defects in EL images. The system fuses the deep feature representations extracted from two different deep learning models (Inception-V3 and ResNet50) to form more discriminative feature vectors. These feature vectors are then fed into the classifier layer to assign them into one of different types of defects. A large-scale, challenging solar cells dataset composed of 2,624 EL images was used to assess the performance of the proposed system in both the binary classification (functional vs defective) task and multi-class classification (functional, mild, moderate, and severe) task. The proposed system has managed to detect the correct defect type with less than 1 s per image with an accuracy rate of 98.15% and 95.35% in the binary classification and multi-classification task, respectively.

Inter-frame video forgery detection using UFS-MSRC algorithm and LSTM network

The forgery involved in region duplication is a common type of video tampering, where the traditional techniques used to detect video tampering are ineffective and inefficient for the forged videos under complex backgrounds. To overcome this issue, a novel video forgery detection model is introduced in this research paper. Initially, the input video sequences are collected from Surrey University Library for Forensic Analysis (SULFA) and Sondos datasets. Further, spatiotemporal averaging method is carried out on the collected video sequences to obtain background information with a pale of moving objects for an effective video forgery detection. Next, feature extraction is performed using the GoogLeNet model for extracting the feature vectors. Then, the Unsupervised Feature Selection with Multi-Subspace Randomization and Collaboration (UFS-MSRC) approach is used to choose the discriminative feature vectors that superiorly reduce the training time and improve the detection accuracy. Finally, long short-term memory (LSTM) network is applied for forgery detection in the different video sequences. The experimental evaluation illustrated that the UFS-MSRC with LSTM model attained 98.13% and 97.38% of accuracy on SULFA and Sondos datasets, where the obtained results are better when compared to the existing models in video forgery detection.

Wearable multi-sensor data fusion approach for human activity recognition using machine learning algorithms

Wearable sensor based human activity recognition (HAR) has a broad range of applications in healthcare, fitness, smart home, and surveillance. In spite of the substantial amount of computational research on HAR, the open challenges in multi-sensor based activity recognition including complex temporal data, identifying discriminative feature vectors from multi-modal data and dimensionality reduction require considerable research attention. Hence, in this paper, using multi-resolution time-frequency analysis of received signal strength (RSS) between the wearable sensors, we present a supervised machine learning (ML) based activity recognition framework. The multi-sensor activity data acquired using the wireless sensor network (WSN) nodes and inertial-sensor embedded in a smartphone are decomposed using the discrete wavelet transform (DWT) and the empirical mode decomposition (EMD) techniques for extracting the prominent feature vector. Using the discriminative statistical features from DWT along with the entropy features from EMD, the four ML classifier models such as support vector machine (SVM), K-nearest neighbour (KNN), ensemble classifier (EC), and decision tree (DT) are trained to classify various human activities. The efficacy of the proposed design is assessed on a publicly available dataset from UCI. The experimental results assessed using the confusion matrix and parallel coordinate plot (PCP) substantiate that the proposed ML based HAR framework can achieve a maximum classification accuracy of 99.63% and is superior to several state-of-the-art ML techniques.

Analysis of Aerobics Auxiliary Training Based on Deep Learning

With the in-depth integration of information technology and subject teaching, it is also an inevitable trend to apply modern information technology to aerobics teaching. In this paper, the N-best algorithm is used in the video and real-time camera in aerobics, so that the human posture parameters in a single-frame image can be estimated. By using the relative position and motion direction of each part of the human body to describe the characteristics of aerobics, the Laplace scoring method is used to reduce the dimension of the data, and the discriminant human motion feature vector with a strong local topological structure is obtained. Finally, the iterative self-organizing data analysis technology (ISODATA algorithm) is used to dynamically determine the keyframe. In the aerobics video keyframe extraction experiments, the ST-FMP model improves the recognition accuracy of nondeterministic body parts of the flexible hybrid articulated human model (FMP) by about 15 percentage points and achieves 81% keyframe extraction accuracy, which is better than the keyframe algorithms of KFE and motion block. The proposed algorithm is sensitive to human motion features and human pose and is suitable for motion video annotation review.

Classification of Improvised Explosive Devices Using Multilevel Projective Dictionary Learning With Low-Rank Prior

Improvised explosive devices (IEDs) pose a significant threat to defense forces and humanitarian demining personnel. They are weapons of modern times, made from nonconventional military materials, rendering them difficult to identify when buried in the ground. Numerous studies focus on detecting these explosive threats and reducing the false alarm rate. However, there are few attempts to identify the detected explosive devices to take proper countermeasures. This article presents a multilevel projective dictionary learning (DL) method to classify ground-penetrating radar signals from IEDs. The proposed dictionary learning method solves three different tasks simultaneously: suppressing background clutter, learning a set of discriminative features for classification, and training a classifier. The suppression of ground clutter is formulated as a low-rank (LR) optimization problem with sparse constraints, where a low-rank subspace is learned from background clutter signals. Dictionary learning is used to transform the target signals into discriminative feature vectors, which are in turn used by the classifier to predict the target class. Experiments were conducted on real radar data. The results showed that the proposed method is more effective than the existing dictionary models and machine learning methods.

A Machine Learning Method for Detection of Surface Defects on Ceramic Tiles Using Convolutional Neural Networks

We propose a simple but effective convolutional neural network to learn the similarities between closely related raw pixel images for feature representation extraction and classification through the initialization of convolutional kernels from learned filter kernels of the network. The binary-class classification of sigmoid and discriminative feature vectors are simultaneously learned together contrasting the handcrafted traditional method of feature extractions, which split feature-extraction and classification tasks into two different processes during training. Relying on the high-quality feature representation learned by the network, the classification tasks can be efficiently conducted. We evaluated the classification performance of our proposed method using a collection of tile surface images consisting of cracked surfaces and no-cracked surfaces. We tried to classify the tiny-cracked surfaces from non-crack normal tile demarcations, which could be useful for automated visual inspections that are labor intensive, risky in high altitudes, and time consuming with manual inspection methods. We performed a series of comparisons on the results obtained by varying the optimization, activation functions, and deployment of different data augmentation methods in our network architecture. By doing this, the effectiveness of the presented model for smooth surface defect classification was explored and determined. Through extensive experimentation, we obtained a promising validation accuracy and minimal loss.

A hybrid deep transfer learning-based approach for Parkinson's disease classification in surface electromyography signals

Parkinson's disease (PD) is known as a rampant neurodegenerative disorder, which has afflicted approximately 10 million people throughout the world. Surface Electromyography (sEMG) signal trials received from the upper extremities, such as the arm and wrist, would be an efficient way to assess neuromuscular function in the detection of PD. This paper mainly aimed to utilize pre-trained deep transfer learning (DTL) structures and conventional machine learning (ML) models as an automated approach to diagnose PD from sEMG signals. Primarily, we stacked the extracted features from three deep pre-trained architectures, including AlexNet, VGG-f, and CaffeNet, to generate the discriminative feature vectors. Although the number of stacked features from all the three deep structures was large, the proper features is effective in overcoming the challenge of over-fitting as well as increasing the robustness to added noise with different levels. Subsequently, we proposed a novel soft combination of subset feature selection methods, including receiver operating characteristic (ROC), entropy, and the signal-to-noise (SNR) procedures, in order to reduce the size of the extracted features. Finally, we utilized the support vector machine (SVM) with radial basis function (RBF) kernel for identifying PD disorder. The experimental results in different analysis frameworks illustrated that the hybrid deep transfer learning-based approach to PD classification could lead to hitting rates higher than 99%. Moreover, it can be of a competitive performance with the state-of-the-art SVM-based pattern even though the suggested model needs minimal processing in feature construction of sEMG signals to PD detection.

Training augmentation with TANDEM acoustic modelling in Punjabi adult speech recognition system

Processing of low resource pre and post acoustic signals always faced the challenge of data scarcity in its training module. It’s difficult to obtain high system accuracy with limited corpora in train set which results into extraction of large discriminative feature vector. These vectors information are distorted due to acoustic mismatch occurs because of real environment and inter speaker variations. In this paper, context independent information of an input speech signal is pre-processed using bottleneck features and later in modeling phase Tandem-NN model has been employ to enhance system accuracy. Later to fulfill the requirement of train data issues, in-domain training augmentation is perform using fusion of original clean and artificially created modified train noisy data and to further boost this training data, tempo modification of input speech signal is perform with maintenance of its spectral envelope and pitch in corresponding input audio signal. Experimental result shows that a relative improvement of 13.53% is achieved in clean and 32.43% in noisy conditions with Tandem-NN system in comparison to that of baseline system respectively.

Balanced multiple weighted linear discriminant analysis and its application to visual process monitoring

Abstract Visual process monitoring is important in complex chemical processes. To address the high state separation of industrial data, we propose a new criterion for feature extraction called balanced multiple weighted linear discriminant analysis (BMWLDA). Then, we combine BMWLDA with self-organizing map (SOM) for visual monitoring of industrial operation processes. BMWLDA can extract the discriminative feature vectors from the original industrial data and maximally separate industrial operation states in the space spanned by these discriminative feature vectors. When the discriminative feature vectors are used as the input to SOM, the training result of SOM can differentiate industrial operation states clearly. This function improves the performance of visual monitoring. Continuous stirred tank reactor is used to verify that the class separation performance of BMWLDA is more effective than that of traditional linear discriminant analysis, approximate pairwise accuracy criterion, max–min distance analysis, maximum margin criterion, and local Fisher discriminant analysis. In addition, the method that combines BMWLDA with SOM can effectively perform visual process monitoring in real time.

DeepFrag-k: a fragment-based deep learning approach for protein fold recognition

BackgroundOne of the most essential problems in structural bioinformatics is protein fold recognition. In this paper, we design a novel deep learning architecture, so-called DeepFrag-k, which identifies fold discriminative features at fragment level to improve the accuracy of protein fold recognition. DeepFrag-k is composed of two stages: the first stage employs a multi-modal Deep Belief Network (DBN) to predict the potential structural fragments given a sequence, represented as a fragment vector, and then the second stage uses a deep convolutional neural network (CNN) to classify the fragment vector into the corresponding fold.ResultsOur results show that DeepFrag-k yields 92.98% accuracy in predicting the top-100 most popular fragments, which can be used to generate discriminative fragment feature vectors to improve protein fold recognition.ConclusionsThere is a set of fragments that can serve as structural “keywords” distinguishing between major protein folds. The deep learning architecture in DeepFrag-k is able to accurately identify these fragments as structure features to improve protein fold recognition.