Structural Feature Extraction Research Articles

Patch-Sorted Deep Feature Learning for High Resolution SAR Image Classification

Synthetic aperture radar (SAR) image classification is a fundamental process for SAR image understanding and interpretation. The traditional SAR classification methods extract shallow and handcrafted features, which cannot subtly depict the abundant modal information in high resolution SAR image. Inspired by deep learning, an effective feature learning tool, a novel method called patch-sorted deep neural network (PSDNN) to implement unsupervised discriminative feature learning is proposed. First, the randomly selected patches are measured and sorted by the meticulously designed patch-sorted strategy, which adopts instance-based prototypes learning. Then the sorted patches are delivered to a well-designed dual-sparse autoencoder to obtain desired weights in each layer. Convolutional neural network is followed to extract high-level spatial and structural features. At last, the features are fed to a linear support vector machine to generate predicted labels. The experimental results in three broad SAR images of different satellites confirm the effectiveness and generalization of our method. Compared with three traditional feature descriptors and four unsupervised deep feature descriptors, the features learned in PSDNN appear powerful discrimination and the PSDNN achieves desired classification accuracy and a good visual appearance.

A Novel Affine and Contrast Invariant Descriptor for Infrared and Visible Image Registration

Infrared and visible image registration is a very challenging task due to the large geometric changes and the significant contrast differences caused by the inconsistent capture conditions. To address this problem, this paper proposes a novel affine and contrast invariant descriptor called maximally stable phase congruency (MSPC), which integrates the affine invariant region extraction with the structural features of images organically. First, to achieve the contrast invariance and ensure the significance of features, we detect feature points using moment ranking analysis and extract structural features via merging phase congruency images in multiple orientations. Then, coarse neighborhoods centered on the feature points are obtained based on Log-Gabor filter responses over scales and orientations. Subsequently, the affine invariant regions of feature points are determined by using maximally stable extremal regions. Finally, structural descriptors are constructed from those regions and the registration can be implemented according to the correspondence of the descriptors. The proposed method has been tested on various infrared and visible pairs acquired by different platforms. Experimental results demonstrate that our method outperforms several state-of-the-art methods in terms of robustness and precision with different image data and also show its effectiveness in the application of trajectory tracking.

Human Brain Biometric Identification Method by Extraction of Brain Print

Identification of a human being through visual structural features face challenge of forgery. The hidden biometrics is comparatively recent yet it has appeared as optimistic modality for security. We believe that the distinguishing feature among human is the brain so qualifying and quantifying the uniqueness in structure of the brain should lead to subject identification. Previously, brain signals have been used for biometrics but like behavioral biometrics these can be deceiving. In this paper, we concentrate on extraction of concrete structural features from human brain which are non-deceivable. Our aim is to extract non-linear curves having approximate brain structural information instead of considering brain to be of predefined abstract regular shapes. The results are optimistic as we are able to extract enough number of brain curves from a securely selected slice from complete brain map ensuring the scalability of the approach.

Automated quantification of renal interstitial fibrosis for computer-aided diagnosis: A comprehensive tissue structure segmentation method

Interstitial fibrosis in renal biopsy samples is a scarring tissue structure that may be visually quantified by pathologists as an indicator to the presence and extent of chronic kidney disease. The standard method of quantification by visual evaluation presents reproducibility issues in the diagnoses. This study proposes an automated quantification system for measuring the amount of interstitial fibrosis in renal biopsy images as a consistent basis of comparison among pathologists. The system extracts and segments the renal tissue structures based on colour information and structural assumptions of the tissue structures. The regions in the biopsy representing the interstitial fibrosis are deduced through the elimination of non-interstitial fibrosis structures from the biopsy area and quantified as a percentage of the total area of the biopsy sample. A ground truth image dataset has been manually prepared by consulting an experienced pathologist for the validation of the segmentation algorithms. The results from experiments involving experienced pathologists have demonstrated a good correlation in quantification result between the automated system and the pathologists’ visual evaluation. Experiments investigating the variability in pathologists also proved the automated quantification error rate to be on par with the average intra-observer variability in pathologists’ quantification. Background and ObjectiveInterstitial fibrosis in renal biopsy samples is a scarring tissue structure that may be visually quantified by pathologists as an indicator to the presence and extent of chronic kidney disease. The standard method of quantification by visual evaluation presents reproducibility issues in the diagnoses due to the uncertainties in human judgement. MethodsAn automated quantification system for accurately measuring the amount of interstitial fibrosis in renal biopsy images is presented as a consistent basis of comparison among pathologists. The system identifies the renal tissue structures through knowledge-based rules employing colour space transformations and structural features extraction from the images. In particular, the renal glomerulus identification is based on a multiscale textural feature analysis and a support vector machine. The regions in the biopsy representing interstitial fibrosis are deduced through the elimination of non-interstitial fibrosis structures from the biopsy area. The experiments conducted evaluate the system in terms of quantification accuracy, intra- and inter-observer variability in visual quantification by pathologists, and the effect introduced by the automated quantification system on the pathologists’ diagnosis. ResultsA 40-image ground truth dataset has been manually prepared by consulting an experienced pathologist for the validation of the segmentation algorithms. The results from experiments involving experienced pathologists have demonstrated an average error of 9 percentage points in quantification result between the automated system and the pathologists’ visual evaluation. Experiments investigating the variability in pathologists involving samples from 70 kidney patients also proved the automated quantification error rate to be on par with the average intra-observer variability in pathologists’ quantification. ConclusionsThe accuracy of the proposed quantification system has been validated with the ground truth dataset and compared against the pathologists’ quantification results. It has been shown that the correlation between different pathologists’ estimation of interstitial fibrosis area has significantly improved, demonstrating the effectiveness of the quantification system as a diagnostic aide.

Evaluating linear geological structures in seismogenic compressional setting, Kashmir basin, NW-Himalaya

Kashmir basin situated in tectonically active area NW Himalaya is surrounded by a number of seismogenic faults. In the present study, the structures are identified using the edge enhancement filters, false colour composite and DEM derived products like shaded relief, aspect, slope, drainage network, and 3D profiling. Linear structural feature map was analysed using density, orientation, length, and subsequently correlated with geological map. The area shows highest density in NW–SE, E–W and N–S orientation, with predominance of first direction structures with trend resemblance to tectonic structures present in and around. The western flank shows high density of tectonic lineaments, the Gulmarg and Tosamaidan lineament are supported by geomorphic evidences and shows linear trend with the tectonic structures including Balapur fault (out-of-sequence fault) mapped in Pleistocene deposits and earthquake epicentres present in and around the area. These results correspond to the overall NW–SE direction of the structures resulting from the NE–SW compressional effect of the regional tectonic structures. The approach applied for extraction of structural features of tectonic origin will prove to be very crucial for further investigation in future for fault delineation, seismic and landslide hazard assessment, and mitigation measures in the region.

A novel nucleic acid sequence encoding strategy for high-performance aptamer identification and the aid of sequence design and optimization

Aptamers have exhibited a great potential for research, clinical and industrial purposes. A critical step to realize these applications is to gain high-affinity aptamers specific to interested targets. To facilitate the selection of aptamers generated in systematic evolution of ligands by exponential enrichment (SELEX) process, we propose a novel nucleic acid sequence encoding strategy of Apta-LoopEnc for secondary structural feature extraction of candidate sequences by analyzing their delicate substructures in loop regions. Since the unique loop structures of aptamers determine their interaction with targets, encoding their central loop structures directly enables featuring aptamer binding affinity related properties. Additionally, the nucleotide composition of a sequence is also used as descriptors in Apta-LoopEnc to further decrease the description similarity between sequences. The feasibility of Apta-LoopEnc for sequence encoding has been demonstrated by the study of high-affinity aptamer identification against human hepatocellular carcinoma cells. The results indicate the developed Apta-LoopEnc is able to significantly improve the performance of different pattern recognition models. Using the Apta-LoopEnc based support vector machine (SVM) to predict a set of newly designed candidate sequences beyond SELEX has further demonstrated the potential of the developed sequence encoding and prediction strategy in aid of high-performance aptamer design and optimization in an easy, time-saving and cost-effective way via computation, thus, promoting the development of aptamer-related studies and applications.

Supervised Classification of Power Lines from Airborne LiDAR Data in Urban Areas

Automatic extraction of power lines using airborne LiDAR (Light Detection and Ranging) data has been one of the most important topics for electric power management. However, this is very challenging over complex urban areas, where power lines are in close proximity to buildings and trees. In this paper, we presented a new, semi-automated and versatile framework that consists of four steps: (i) power line candidate point filtering, (ii) local neighborhood selection, (iii) spatial structural feature extraction, and (iv) SVM classification. We introduced the power line corridor direction for candidate point filtering and multi-scale slant cylindrical neighborhood for spatial structural features extraction. In a detailed evaluation involving seven scales and four types for local neighborhood selection, 26 structural features, and two datasets, we demonstrated that the use of multi-scale slant cylindrical neighborhood for individual 3D points significantly improved the power line classification. The experiments indicated that precision, recall and quality rate of power line classification is more than 98%, 98% and 97%, respectively. Additionally, we showed that our approach can reduce the whole processing time while achieving high accuracy.

Perceptual hashing for color images based on hybrid extraction of structural features

In this paper, a novel perceptual hashing scheme for color images is proposed with the hybrid feature extraction mechanism. During the stage of pre-processing, image normalization, Gaussian low-pass filtering and singular value decomposition (SVD) are applied on original image to improve the robustness of the scheme. In order to fully extract the structural features, the circle-based and the block-based strategies are exploited to sample the salient edge points with the aid of Canny operator, and then the color vector angles, which can effectively describe color pixel information, are calculated for the sampled points. Finally, after quantizing and scrambling the variances of the color vector angles for these sampled points, the image hash can be generated securely. Experimental results demonstrate that our scheme can achieve satisfactory performances with respect to perceptual robustness and discrimination.

Deep learning-based subdivision approach for large scale macromolecules structure recovery from electron cryo tomograms.

MotivationCellular Electron CryoTomography (CECT) enables 3D visualization of cellular organization at near-native state and in sub-molecular resolution, making it a powerful tool for analyzing structures of macromolecular complexes and their spatial organizations inside single cells. However, high degree of structural complexity together with practical imaging limitations makes the systematic de novo discovery of structures within cells challenging. It would likely require averaging and classifying millions of subtomograms potentially containing hundreds of highly heterogeneous structural classes. Although it is no longer difficult to acquire CECT data containing such amount of subtomograms due to advances in data acquisition automation, existing computational approaches have very limited scalability or discrimination ability, making them incapable of processing such amount of data.ResultsTo complement existing approaches, in this article we propose a new approach for subdividing subtomograms into smaller but relatively homogeneous subsets. The structures in these subsets can then be separately recovered using existing computation intensive methods. Our approach is based on supervised structural feature extraction using deep learning, in combination with unsupervised clustering and reference-free classification. Our experiments show that, compared with existing unsupervised rotation invariant feature and pose-normalization based approaches, our new approach achieves significant improvements in both discrimination ability and scalability. More importantly, our new approach is able to discover new structural classes and recover structures that do not exist in training data.Availability and ImplementationSource code freely available at http://www.cs.cmu.edu/∼mxu1/software.Supplementary information Supplementary data are available at Bioinformatics online.

Multiple metric learning based on bar-shape descriptor for person re-identification

The robust structural feature extraction and similarity measure play critical roles in person re-identification. This paper presents a novel algorithm named Multiple Metric Learning based on Bar-shape Descriptor (MMLBD) for person re-identification. Specifically, we first propose a new Multiple Bar-shape Descriptor that can take full account of the spatial correlation between the center points and their adjacent points on different directions. It captures further histogram features based on a novel color difference weight factors with an overlapping sliding window, which can depict the local variations and consistency in the whole image. The similarity and dissimilarity of samples are used to train the weight factor of features and an optimal subspace could be obtained at the same time. Next, we provide an effective multiple metric learning method fusing two-channel bar-shape structural features via the optimal similarity pairwise measure obtained by a dissimilarity matrix. This measure can fully mine the discriminative information and eliminate redundancy in the similar features, which make the MMLBD simple and effective. Finally, evaluation experiments on the i_LIDS, CAVIAR4REID and WARD data-sets are carried out, which compare the proposed MMLBD with the corresponding methods. Experimental results demonstrate that the MMLBD is more effective and robust against visual appearance variations.

NEAT-FLEX: Predicting the conformational flexibility of amino acids using neuroevolution of augmenting topologies.

The development of computational methods to accurately model three-dimensional protein structures from sequences of amino acid residues is becoming increasingly important to the structural biology field. This paper addresses the challenge of predicting the tertiary structure of a given amino acid sequence, which has been reported to belong to the NP-Complete class of problems. We present a new method, namely NEAT-FLEX, based on NeuroEvolution of Augmenting Topologies (NEAT) to extract structural features from (ABS) proteins that are determined experimentally. The proposed method manipulates structural information from the Protein Data Bank (PDB) and predicts the conformational flexibility (FLEX) of residues of a target amino acid sequence. This information may be used in three-dimensional structure prediction approaches as a way to reduce the conformational search space. The proposed method was tested with 24 different amino acid sequences. Evolving neural networks were compared against a traditional error back-propagation algorithm; results show that the proposed method is a powerful way to extract and represent structural information from protein molecules that are determined experimentally.

Low-level structure feature extraction for image processing via stacked sparse denoising autoencoder

In this paper, we propose a novel low-level structure feature extraction for image processing based on deep neural network, stacked sparse denoising autoencoder (SSDA). The current image processing methods via deep learning are directly building and learning the end-to-end mappings between the input/output. Instead, we advocate the analysis of the first layer learning features from input data. With the learned low-level structure features, we improve two edge-preserving filters that are key to image processing tasks such as denoising, High Dynamic Range (HDR) compression and details enhancement. Due to the validity and superiority of the proposed feature extraction, the results computed by the two improved filters do not suffer from the drawbacks including halos, edge blurring, noise amplification and over-enhancement. More importantly, we demonstrate that the features trained from natural images are not specific and can extract the structure features of infrared images. Hence, it is feasible to handle tasks by using the trained features directly.

ALDOCX: Detection of Unknown Malicious Microsoft Office Documents Using Designated Active Learning Methods Based on New Structural Feature Extraction Methodology

Attackers increasingly take advantage of innocent users who tend to casually open email messages assumed to be benign, carrying malicious documents. Recent targeted attacks aimed at organizations utilize the new Microsoft Word documents (*.docx). Anti-virus software fails to detect new unknown malicious files, including malicious docx files. In this paper, we present ALDOCX, a framework aimed at accurate detection of new unknown malicious docx files that also efficiently enhances the framework’s detection capabilities over time. Detection relies upon our new structural feature extraction methodology (SFEM), which is performed statically using meta-features extracted from docx files. Using machine-learning algorithms with SFEM, we created a detection model that successfully detects new unknown malicious docx files. In addition, because it is crucial to maintain the detection model’s updatability and incorporate new malicious files created daily, ALDOCX integrates our active-learning (AL) methods, which are designed to efficiently assist anti-virus vendors by better focusing their experts’ analytical efforts and enhance detection capability. ALDOCX identifies and acquires new docx files that are most likely malicious, as well as informative benign files. These files are used for enhancing the knowledge stores of both the detection model and the anti-virus software. The evaluation results show that by using ALDOCX and SFEM, we achieved a high detection rate of malicious docx files (94.44% TPR) compared with the anti-virus software (85.9% TPR)—with very low FPR rates (0.19%). ALDOCX’s AL methods used only 14% of the labeled docx files, which led to a reduction of 95.5% in security experts’ labeling efforts compared with the passive learning and the support vector machine (SVM)-Margin (existing active-learning method). Our AL methods also showed a significant improvement of 91% in number of unknown docx malware acquired, compared with the passive learning and the SVM-Margin, thus providing an improved updating solution for the detection model, as well as the anti-virus software widely used within organizations.

Coffee plantation area recognition in satellite images using Fourier transform

In this study, a machine vision scheme is proposed for coffee plantation area recognition in satellite images. It automatically segments the row-planted coffee field from forest trees and irrelevant areas in the image. The result can be used for coffee yield estimation to improve the supply and demand of coffee commodity in the market. Commercial coffee plantation grows coffee trees in rows along a specific direction to increase the production yield and management efficiency. The coffee plants and forest trees present the same color tone in the image and, thus, color cannot be used for the discrimination. The row-planting pattern of coffee trees shows structural texture in the satellite image. This study presents a Fourier transform-based method to extract structural features in the spectral domain for image segmentation. Row-planted coffee fields generate high-energy frequency components in a single direction, while naturally-growing plants present omnidirectional frequency components in the spectral domain image. The main frequency in the power spectrum indicates the number of parallel lines in a small patch window and, thus, gives the density feature. The density feature for the row-planted coffee filed is equivalent to the number of rows in a unit square area, whereas it is only one for the randomly-growing plants. This study analyzes the satellite images of coffee plantation regions in different times with varying illuminations and growing stages in Brazil, Africa, Vietnam and Hawaii. The experimental results have shown that the Fourier-based structural and density features can provide correct segmentation to distinguish the row-planted coffee field from irrelevant vegetation areas in the satellite image.

Structure similarity-guided image binarization for automatic segmentation of epidermis surface microstructure images

Partitioning epidermis surface microstructure (ESM) images into skin ridge and skin furrow regions is an important preprocessing step before quantitative analyses on ESM images. Binarization segmentation is a potential technique for partitioning ESM images because of its computational simplicity and ease of implementation. However, even for some state-of-the-art binarization methods, it remains a challenge to automatically segment ESM images, because the grey-level histograms of ESM images have no obvious external features to guide automatic assessment of appropriate thresholds. Inspired by human visual perceptual functions of structural feature extraction and comparison, we propose a structure similarity-guided image binarization method. The proposed method seeks for the binary image that best approximates the input ESM image in terms of structural features. The proposed method is validated by comparing it with two recently developed automatic binarization techniques as well as a manual binarization method on 20 synthetic noisy images and 30 ESM images. The experimental results show: (1) the proposed method possesses self-adaption ability to cope with different images with same grey-level histogram; (2) compared to two automatic binarization techniques, the proposed method significantly improves average accuracy in segmenting ESM images with an acceptable decrease in computational efficiency; (3) and the proposed method is applicable for segmenting practical EMS images. (Matlab code of the proposed method can be obtained by contacting with the corresponding author.).

Odia character recognition using backpropagation network with binary features

Automatic recognition of printed characters has been an area of active research since last few decades. Though considerable work has been reported for Latin, CJK, and many other popular languages, but a few works has been reported for many Indian languages. In this work, we propose a complete model with effective techniques to recognise printed offline Odia language characters. First we develop a technique to segment the image to extract characters, and then features of these segmented characters are extracted using a binary feature extraction, as well as structural feature extraction. Finally, these extracted features are used to classify character symbols using a modified backpropagation network. In each of these phases it is found that the proposed technique outperforms existing techniques and yields high accuracy rate. Further, this work has tried to include all the possible characters (approx. 2,500 characters) in training/testing unlike other works, where a small set of characters are normally considered for testing.

Analysis of confocal microscopy image data of Physcomitrella chloroplasts to reveal adaptation principles leading to structural stability at the nanoscale

AbstractProteins of the FtsZ (filamentous temperature‐sensitive Z) family establish complex polymeric spatial patterns in plastids of the moss Physcomitrella patens. These structures represent a "plastoskeleton" that might contribute to plastid shape and stability. The aim of this work is to develop computer models of FtsZ network connectivity and dynamics. These computer models are developed from green fluorescent protein (GFP) images obtained by Prof. Reski's Plant Biotechnology group at the University of Freiburg by means of laser scanning confocal microscopy. Subsequently, structural reconstruction techniques based on cellular automata algorithm are applied on the raw image data to compensate for any information that is lost in the process of imaging. Segmentation of the processed data will result in the computer model of the targeted structures, which are used for further analysis. Model generation and experimental testing in the moss system is conducted in an iterative process with the aim of both forward and reverse biomimetics. A developed image processing algorithm with the aim of quantification of structural characteristics is utilized to extract structural features of the FtsZ network in Physcomitrella enabling us a statistical analysis on the extracted data. (© 2016 Wiley‐VCH Verlag GmbH & Co. KGaA, Weinheim)

Sports Video Structure Analysis and Feature Extraction in Long Jump Video

How to help people to find their favorite sports in the massive video, in order to achieve this goal in the finite state machine (FSM) theory based on in-depth research. Combined with the actual situation of this paper, first, we determined the reasonable FSM model. Then, a fast robust global motion estimation algorithm is used to estimate the global motion of the video sequences, and the foreground is separated from the background by motion compensation. After the foreground of the video, a series of image feature extraction algorithm, which is based on the main color histogram and histogram, is used to extract the low level feature extraction. Finally, the results of the experimental operation, the performance of the algorithm, and the results of the system are presented.

SFEM: Structural feature extraction methodology for the detection of malicious office documents using machine learning methods

Office documents are used extensively by individuals and organizations. Most users consider these documents safe for use. Unfortunately, Office documents can contain malicious components and perform harmful operations. Attackers increasingly take advantage of naive users and leverage Office documents in order to launch sophisticated advanced persistent threat (APT) and ransomware attacks. Recently, targeted cyber-attacks against organizations have been initiated with emails containing malicious attachments. Since most email servers do not allow the attachment of executable files to emails, attackers prefer to use of non-executable files (e.g., documents) for malicious purposes. Existing anti-virus engines primarily use signature-based detection methods, and therefore fail to detect new unknown malicious code which has been embedded in an Office document. Machine learning methods have been shown to be effective at detecting known and unknown malware in various domains, however, to the best of our knowledge, machine learning methods have not been used for the detection of malicious XML-based Office documents (*.docx, *.xlsx, *.pptx, *.odt, *.ods, etc.). In this paper we present a novel structural feature extraction methodology (SFEM) for XML-based Office documents. SFEM extracts discriminative features from documents, based on their structure. We leveraged SFEM’s features with machine learning algorithms for effective detection of malicious *.docx documents. We extensively evaluated SFEM with machine learning classifiers using a representative collection (16,938 *.docx documents collected "from the wild") which contains ∼4.9% malicious and ∼95.1% benign documents. We examined 1,600 unique configurations based on different combinations of feature extraction, feature selection, feature representation, top-feature selection methods, and machine learning classifiers. The results show that machine learning algorithms trained on features provided by SFEM successfully detect new unknown malicious *.docx documents. The Random Forest classifier achieves the highest detection rates, with an AUC of 99.12% and true positive rate (TPR) of 97% that is accompanied by a false positive rate (FPR) of 4.9%. In comparison, the best anti-virus engine achieves a TPR which is ∼25% lower.

Tifinagh Printed Character Recognition through Structural Feature Extraction

In this paper a system for the recognition of printed Tifinagh characters is presented. It is divided into three main steps: preprocessing, feature extraction, and classification. Image quality is enhanced through preprocessing which are: binarization, normalization and thinning. Then the image is given to a proposed structural feature extracting algorithm where the character is divided into several geometrically sample shapes which are segments, then transformed into an undirected graph with unique coordinate of all nodes. The character is classified by matching the graph of the character and its counterpart graph which is generated from the images in the IRCAM database using an efficient spectral graph matching algorithm. Experimental results and analysis are accomplished by the use of 3267 random characters to test the effectiveness. The system shows good results in term of accuracy and CPU time.