Feature Vector Size Research Articles

Color Image Retrieval Using Compacted Feature Vector with Mean-Count Tree

Image Retrieval systems highly rely on the image signatures stored in database. Constructing Image signatures with optimal size and accurate representation of image is most interesting challenge here. High level image features like object and their characteristic are useful in sentiment image analysis but faces the limitations of domain specific feature vector. Low level image features like color, texture and shape are interesting and useful enough to represent the image in diverse image databases. Targeting to low level image feature: color, the image signatures created are of huge size, as those represents three color planes and their values. Image signatures vary, as the image size varies. Targeting towards creating the optimal image signatures with color feature, to reduce the size of feature vector is possible by considering images in frequency domain. Image transforms converts image in frequency domain, with compressed image data. This data is further reduced form by ignoring the low energy components. This paper discusses the approach to construct image signatures by considering high energy components of transformed image. Further the high energy components are bagged together. Here the intelligent mean-count tree is created based on image information. Performance of Image retrieval is tested using image feature database.

A Novel Approach for Ontology-Based Feature Vector Generation for Web Text Document Classification

The task of extracting the used feature vector in mining tasks (classification, clustering …etc.) is considered the most important task for enhancing the text processing capabilities. This paper proposes a novel approach to be used in building the feature vector used in web text document classification process; adding semantics in the generated feature vector. This approach is based on utilizing the benefit of the hierarchal structure of the WordNet ontology, to eliminate meaningless words from the generated feature vector that has no semantic relation with any of WordNet lexical categories; this leads to the reduction of the feature vector size without losing information on the text, also enriching the feature vector by concatenating each word with its corresponding WordNet lexical category. For mining tasks, the Vector Space Model (VSM) is used to represent text documents and the Term Frequency Inverse Document Frequency (TFIDF) is used as a term weighting technique. The proposed ontology based approach was evaluated against the Principal component analysis (PCA) approach, and against an ontology based reduction technique without the process of adding semantics to the generated feature vector using several experiments with five different classifiers (SVM, JRIP, J48, Naive-Bayes, and kNN). The experimental results reveal the effectiveness of the authors' proposed approach against other traditional approaches to achieve a better classification accuracy F-measure, precision, and recall.

Comparative Study of Local Binary Pattern Derivatives for Low Size Feature Vector Representation in Face Recognition

Comparative Study of Local Binary Pattern Derivatives for Low Size Feature Vector Representation in Face Recognition

Object recognition of real targets using modelled SAR images

In this work the problem of recognition is studied using SAR images. The algorithm of recognition is based on the computation of conjugation indices with vectors of class. The support subspaces for each class are constructed by exception of the most and the less correlated vectors in a class. In the study we examine the ability of a significant feature vector size reduce that leads to recognition time decrease. The images of targets form the feature vectors that are transformed using pre-trained convolutional neural network (CNN).

A Novel Approach for Ontology-Based Dimensionality Reduction for Web Text Document Classification

Dimensionality reduction of feature vector size plays a vital role in enhancing the text processing capabilities; it aims in reducing the size of the feature vector used in the mining tasks (classification, clustering, etc.). This paper proposes an efficient approach to be used in reducing the size of the feature vector for web text document classification process. This approach is based on using WordNet ontology, utilizing the benefit of its hierarchal structure, to eliminate words from the generated feature vector that has no relation with any of WordNet lexical categories; this leads to the reduction of the feature vector size without losing information on the text. For mining tasks, the Vector Space Model (VSM) is used to represent text documents and the Term Frequency Inverse Document Frequency (TFIDF) is used as a term weighting method. The proposed ontology based approach was evaluated against the Principal component analysis (PCA) approach using several experiments. The experimental results reveal the effectiveness of the authors' proposed approach against other traditional approaches to achieve a better classification accuracy F-measure, precision, and recall.

Gait-based Human identification using acoustic sensor and deep neural network

This paper proposes a simple, fast and low-cost gait-based human identification system jointly employing an acoustic sensor system and a deep neural network (DNN) based algorithm to process acoustic data for human identification. The acoustic sensor system is employed to detect and acquire individuals’ gait signatures. Compared with the conventional radar based and video/image based human gait identification technologies, our acoustic sensor system is low-cost and feasible in some special cases, such as in private rooms/hotels or in the night time. We perform identification using a DNN which consists of an auto-encoder and a feed-forward neural network. The experimental results demonstrate that the integrated system of acoustic sensor and DNN considerably improves the system’s performance, which achieves an excellent accuracy rate of 97% for human identification. Aside from the high accuracy of the gait-based human identification, the novel system we developed is also capable of identifying human actions (walk, jog, run, jump, etc.) and the human number which has not been addressed in the literature, thereby significantly enhancing the wide adaptability of our identification system in various applications. Additionally, to better enhance the performance of the proposed human identification method, a large feature set, variable-length frame size and variable-dimension feature vectors are exploited in our identification system. The presented work provides prospects in developing high-performance identification system for large-scale security applications via a simple and efficient technique.

Multi-Class Disease Classification in Brain MRIs Using a Computer-Aided Diagnostic System

Background: An accurate and automatic computer-aided multi-class decision support system to classify the magnetic resonance imaging (MRI) scans of the human brain as normal, Alzheimer, AIDS, cerebral calcinosis, glioma, or metastatic, which helps the radiologists to diagnose the disease in brain MRIs is created. Methods: The performance of the proposed system is validated by using benchmark MRI datasets (OASIS and Harvard) of 310 patients. Master features of the images are extracted using a fast discrete wavelet transform (DWT), then these discriminative features are further analysed by principal component analysis (PCA). Different subset sizes of principal feature vectors are provided to five different decision models. The classification models include the J48 decision tree, k-nearest neighbour (kNN), random forest (RF), and least-squares support vector machine (LS-SVM) with polynomial and radial basis kernels. Results: The RF-based classifier outperformed among all compared decision models and achieved an average accuracy of 96% with 4% standard deviation, and an area under the receiver operating characteristic (ROC) curve of 99%. LS-SVM (RBF) also shows promising results (i.e., 89% accuracy) when the least number of principal features was used. Furthermore, the performance of each classifier on different subset sizes of principal features was (80%–96%) for most performance metrics. Conclusion: The presented medical decision support system demonstrates the potential proof for accurate multi-class classification of brain abnormalities; therefore, it has a potential to use as a diagnostic tool for the medical practitioners.

Arabic document layout analysis

Document layout analysis is a key step in the process of converting document images into text. Arabic language script is cursive and written in different styles which cause some challenges in the analysis of Arabic text documents. In this paper, we introduce an approach for Arabic documents layout analysis. In that approach, the document is segmented into set of zones using morphological operations. The segmented zones are classified as text or non-text ones using a support vector machine classifier. Features used in zone classification are combination between texture-based features and connected component-based features. The textural-based feature vector size is reduced using genetic algorithm. Classified text zones are clustered, using adaptive sample set clustering algorithm, into lines. Each segmented line is segmented into words by clustering inter- and intra-spaces. The proposed system was evaluated against two other systems that represent the best available tools for the Arabic documents analysis, and evaluation results show that the proposed system works well on multi-font and multi-size documents with a variety of layouts even on some historical documents.

Word Embedding Models for Finding Semantic Relationship between Words in Tamil Language

Objective: Word embedding models were most predominantly used in many of the NLP tasks such as document classification, author identification, story understanding etc. In this paper we make a comparison of two Word embedding models for semantic similarity in Tamil language. Each of those two models has its own way of predicting relationship between words in a corpus. Method/Analysis: The term Word embedding in Natural Language Processing is a representation of words in terms of vectors. Word embedding is used as an unsupervised approach instead of traditional way of feature extraction. Word embedding models uses neural networks to generate numerical representation for the given words. In order to find the best model that captures semantic relationship between words, using a morphologically rich language like Tamil would be great. Tamil language is one of the oldest Dravidian languages and it is known for its morphological richness. In Tamil language it is possible to construct 10,000 words from a single root word. Findings: Here we make comparison of Content based Word embedding and Context based Word embedding models respectively. We tried different feature vector sizes for the same word to comment on the accuracy of the models for semantic similarity. Novelty/Improvement: Analysing Word embedding models for morphologically rich language like Tamil helps us to classify the words better based on its semantics.

Effective pixel classification of Mars images based on ant colony optimization feature selection and extreme learning machine

one of the most important tasks of Mars rover, a robot which explores the Mars surface, is the process of automatic segmentation of images taken by front-line Panoramic Camera (Pancam). This procedure is highly significant since the transformation cost of images from Mars to earth is extremely high. Also, image analysis may help Mars rover for its navigation and localization. In this paper, a new feature vector including wavelet and color features for Mars images is proposed. Then, this feature vector is presented for extreme learning machine (ELM) classifier which leads to a high accuracy pixel classifier. It is shown that this system statistically outperforms support vector machine (SVM) and k-nearest neighbours (KNNs) classifiers with respect to both accuracy and run time. After that, dimension reduction in feature space is done by two proposed feature section algorithms based on ant colony optimization (ACO) to decrease the time complexity which is very important in Mars on-board applications. In the first proposed feature selection algorithm, the same feature subset is selected among the feature vector for all pixel classes, while in the second proposed algorithm, the most significant features are selected for each pixel class, separately. Proposed pixel classifier with complete feature set outperforms prior methods by 6.44% and 5.84% with respect to average Fmeasure and accuracy, respectively. Finally, proposed feature selection methods decrease the feature vector size up to 76% and achieves Fmeasure and accuracy of 91.72% and 91.05%, respectively, which outperforms prior methods with 87.22% and 86.64%.

A passive authentication scheme for copy-move forgery based on package clustering algorithm

Copy-move forgery as one of popular methods is widely used to tamper digital images. Passive authentication is extensively used to detect the copy-move forgery images. This paper proposes a passive authentication scheme for copy-move forgery based on the discrete cosine transform (DCT) and the package clustering algorithm. The copy regions and paste regions can be automatically detected in doctored digital images. This scheme works by first applying the DCT to small fixed image blocks to obtain their features and the size of feature vectors are reduced. Moreover, the package clustering algorithm is applied to replace the general lexicographic order technologies to improve the detection precision. The similar blocks can be found by comparing the feature vectors in each package. The experimental results represent that the proposed scheme can locate irregular and meaningful tampered regions and multiply duplicated regions in a suspicious image. The duplicated regions can also be located in digital images that are distorted by adding white Gaussian noise, Gaussian blurring and their mixed operations.

Iris Recognition using Self Mutated Hybrid Wavelet Transform using Cosine, Haar, Hartley and Slant Transforms with Partial Energies of Transformed Iris Images

paper presents iris recognition technique based on the concept of Energy Compaction. Compaction is done by using the partial energies of transformed iris images. Various Self mutated hybrid wavelet transforms (SMHWT) namely Cosine-Haar, Cosine-Hartley, and Cosine-Slant are used to generate feature vector of iris images. The important task is reducing the size of feature vector so that the performance of system can be increased. The reduction of size of feature vector is achieved by using the concept of partial energies. The size of feature vector reduces immensely for 99%, 98%, 97% and 96% of energy. The size of feature vector is extremely large while considering all the coefficients of transformed iris images for 100% of energy which leads more computation. System gives better performance when partial energies are considered. The proposed technique is tested on Palacky University Iris Database. Genuine Acceptance Rate (GAR) is used as a metric to test the performance of the proposed Iris recognition technique. The self mutated hybrid wavelet transform of Cosine-Haar gives preeminent GAR value as compared to other transforms considered. Results show the proposed technique gives better performance in terms of accuracy by considering partial energies as compared to 100% energy.

Boosting local texture descriptors with Log-Gabor filters response for improved image retrieval

In the recent past, many local texture descriptors have been proposed for the image retrieval task. In order to improve the image retrieval accuracy, quite a few of these descriptors have been implemented on Gabor filter response. However, the response of Log-Gabor filters has been proved to be better than Gabor filters with respect to their discrimination ability. In this paper, we present a framework for image retrieval that applies various local texture descriptors on Log-Gabor filters response. To evaluate the retrieval performance of the proposed framework, experiments have been conducted on standard Wang, VisTex and OT-Scene databases. Consistent improvement in the image retrieval accuracy demonstrates the effectiveness of this framework. Further, the experimental results show that the use of proposed framework with low-dimension texture descriptors such as Orthogonal Combination of Local Binary Pattern makes them a better choice over Local Binary Pattern and its high-dimensional variants when higher retrieval accuracy, small feature vector size and ease of computation is desired.

Dimensionality reduction of Fisher vectors for human action recognition

Automatic analysis of human behaviour in large collections of videos is rapidly gaining interest, even more so with the advent of file sharing sites such as YouTube. From one perspective, it can be observed that the size of feature vectors used for human action recognition from videos has been increasing enormously in the last five years, in the order of ∼100–500K. One possible reason might be the growing number of action classes/videos and hence the requirement of discriminating features (that usually end up to be higher-dimensional for larger databases). In this study, the authors review and investigate feature projection as a means to reduce the dimensions of the high-dimensional feature vectors and show their effectiveness in terms of performance. They hypothesise that dimensionality reduction techniques often unearth latent structures in the feature space and are effective in applications such as the fusion of high-dimensional features of different types; and action recognition in untrimmed videos. They conduct all the authors’ experiments using a Bag-of-Words framework for consistency and results are presented on large class benchmark databases such as the HMDB51 and UCF101 datasets.

Content Based Image Retrieval Using Color Mean with Feature Classification Using Naïve Bayes

With the evolution of the Internet, and the availability of image capturing devices such as advanced cameras, picture scanners, the size of advanced image accumulation is expanding quickly. Efficient image searching, browsing and retrieval devices are required by clients from different domains, including remote sensing, fashion, wrongdoing prevention, publishing, medicine, architecture, etc. Here we are extracting color mean features and color standard deviation feature with the proposed method consists of HMMD (Hue Min Max Difference) color plane. It is proved in research work that HMMD along with color mean features and color standard deviation feature is tend to reduced the size of feature vectors, storage space and gives high performance than, RGB-color mean feature. Further, HMMD color space model will be used to improve the feature extraction and improve the precision. At the end, results are presented to show the efficiency of the proposed method. Keywords: HMMD, CBIR, Feature Extraction, Precision, Recall.

A Novel Brain Computer Interface Based on Principle Component Analysis

The main issue to build applicable Brain-Computer Interfaces is the capability to classify the Electroencephalograms (EEG). During the last decade, researchers developed lots of interests in this field. The purpose behind this research is to improve a model for EEG signals analysis. The purpose behind this research is to improve a model for brain signals analysis. We have used high pass filter to remove artifacts, discrete wavelet transform algorithms for feature extraction and statistical features like Mean Absolute Value, Root Mean Square, and Simple Square Integral are used, also we have used principle component analysis to reduce the size of feature vector. It has been depicted from results that the proposed integrated techniques outperform a better performance than methods mentioned in literature.

Bearing Fault Diagnosis Using Feature Ranking Methods and Fault Identification Algorithms

Diagnoses of bearing faults are important to avoid catastrophic failures in rotating machines. This paper presents a methodology to detect various bearing faults from the measured vibration signal. Features such as kurtosis, skewness, mean, root mean square and complexity measure such as Shannon entropy are calculated from time domain,frequency domain and discrete wavelet transform. In total 40 features are calculated from bearing conditions such as Healthy bearing, Inner race fault, Outer race fault and Ball fault. Feature ranking methods such as Chisquare, ReliefF method are used to select most informative feature and subsequently to reduce size of feature vector. Comparison has been made between feature ranking methods and classifiers to obtain best diagnosis result with reduce feature set. Our results shows good fault identification accuracy with minimum number of features.

Score Level Fusion Based Bimodal Biometric Identification Using Thepade's Sorted n-ary Block Truncation Coding with Variod Proportions of Iris and Palmprint Traits

There is increasing interest in the development of reliable, rapid and non-intrusive security control systems. Multimodal biometrics such as palmprints and iris provide highly effective automatic mechanisms for use in personal identification and helps to minimize the system error rate. This paper presents a new method for extracting features in spatial domain from palmprints and iris. Thepade's Block Truncation Coding using level 2 is taken here to reduce the feature vector size. For improving accuracy in form of genuine acceptance rate(GAR)in multimodal biometric identification we take iris and palmprint together at matching score level. The test beds of 60 pairs of Iris and Palmprint samples of 10 person (6 per person of iris as well as Palmprint) are used as test bed for experimentation. In this Paper different color spaces are considered on iris images for improvement in genuine acceptance ratio (GAR). Experimental results consider different matching score proportion of Iris: Palmprint. Using propose techniques with Iris: Palmprint Score 15:1 using TSTBTC Level2 given better performance as indicated by higher GAR values than all other considers scores. RGB color spaces for multimodal fusion of Iris: Palmprint gives high GAR than all other color spaces like YCgCb, YIQ, YCbCr, YIQ, and KLUV.

Periocular localization and feature extraction for human recognition

The advent of biometric system as a next-generation solution towards bringing social and national security to a technically-achievable scenario. This paradigm of authentication has easily taken over the classical token-based and knowledge-based systems. The last decade has seen researches claiming face and iris to be the most promising two traits. Iris produces high accuracy with extremely high-resolution near-infrared (NIR) images, and face is capable of producing moderate accuracy even from low resolution images. To bridge the gap between these two, periocular (periphery of ocular) biometric came into highlight and researchers have initially established its ability to yield accurate recognition. This thesis attempts to design a periocular biometric system. Periocular region can be considered as the region around eye where features, that can participate in uniquely identifying an individual, are existing. So, starting from eye, while moving away from eye, periocular region ranges up to the portion where the skin becomes smooth and no feature is available. Hence periocular biometric, unlike most common segmentation application, cannot be localized through edge detection. The first part of the thesis investigates to identify four trait-specific localization techniques. For achieving perfect localization, (a) conformation of the localization to human anthropometry, (b) high accuracy from a localized image, (c) conformation to human judgement, and (d) subdivision of eye portion are done. The second part concentrates to design a suitable feature extraction method for periocular biometric. The thesis presents a novel Phase Intensive Global Pattern (PIGP) which is shown to able to extract gross as well as subtle features and work well for images without rotation. The next part of the thesis incorporates and ensures scale-invariant and rotation-invariant properties into PIGP, and this modified version is termed as Phase Intensive Local Pattern (PILP). PILP is experimentally proven to work well for NIR databases as well as visual-spectrum (VS) databases. Ability of PILP to identify large number of potential keypoints and extraction of high-dimensional (128D) feature from them results into the high accurate performance of PILP. However, this type of phase-difference based keypoint detection and oriented histogram based large feature extraction is extremely time-consuming and the feature vector, being so large, invites a reduction technique to be employed. The next part of the thesis hence develops a post-reduction technique to reduce the feature vector size and thereby the matching time. Reduced PILP (R-PILP) is developed from PILP by classifying keypoints through verifying the degree of monotonic nature in them. Experiments show that R-PILP is a little less accurate than PILP but R-PILP is faster as compared. All results in the thesis have been derived on four standard publicly available databases: BATH and CASIAv3 (NIR databases), and UBIRISv2 and FERETv4 (VS databases). Comparative analysis have been made with existing landmark techniques like Circular Local Binary Pattern (CLBP), Walsh Mask, Scale Invariant Feature Transform (SIFT), and Speeded Up Robust Features (SURF). It has been observed that these features consistently work equally well on NIR databases. However, performance of existing techniques degrade rapidly when experimented on VS databases. Though the proposed techniques suffers degradation, but outperforms the existing techniques with a high margin. The localization technique, and three progressively developed features PIGP, PILP, and R-PILP complete the objective of developing the periocular biometric system.

Feature extraction from smartphone inertial signals for human activity segmentation

This paper proposes the adaptation of well-known strategies successfully used in speech processing: Mel Frequency Cepstral Coefficients (MFCCs) and Perceptual Linear Prediction (PLP) coefficients. Additionally characteristics like RASTA filtering or delta coefficients are also considered and evaluated for inertial signal processing. These adaptations have been incorporated into a Human Activity Recognition and Segmentation (HARS) system based on Hidden Markov Models (HMMs) for recognizing and segmenting six different physical activities: walking, walking–upstairs, walking-downstairs, sitting, standing and lying.All experiments have been done using a publicly available dataset named UCI Human Activity Recognition Using Smartphones, which includes several sessions with physical activity sequences from 30 volunteers. This dataset has been randomly divided into six subsets for performing a six-fold cross validation procedure. For every experiment, average values from the six-fold cross-validation procedure are shown.The results presented in this paper overcome significantly baseline error rates, constituting a relevant contribution in the field. Adapted MFCC and PLP coefficients improve human activity recognition and segmentation accuracies while reducing feature vector size considerably. RASTA-filtering and delta coefficients contribute significantly to reduce the segmentation error rate obtaining the best results: an Activity Segmentation Error Rate lower than 0.5%.