Fisher Linear Discriminant Classifier Research Articles

Feature reduction of rich features for universal steganalysis using a metaheuristic approach

The development of content adaptive steganographies has become a challenge for steganalysis. This led researchers towards extraction of a rich space of features. The detection of stego images based on spatial rich model (SRM) features and its variants is a promising research area in the field of universal steganalysis. SRM features are extracted as 106 sub-models which collectively provide 34,671 features. So, one of the most significant challenges in universal steganalysis is feature selection. In this paper an improved binary particle swarm optimisation, global and local best particle swarm optimisation (GLBPSO) with Fisher linear discriminant classifier is used to identify relevant feature sub-models which improve the efficiency of a steganalyser. The significant reduction rate of more than 70% is achieved by the proposed approach. This further helps in reducing computational complexity without much affecting the detection capability. The proposed methodology gives superior results when compared with state-of-the-art algorithms.

Designing, manufacturing, and evaluating the diagnostic system of carob moth in pomegranate fruit using digital signal processing

Sound analysis is one of the least destructive testing methods that have been developed for the determination of fruit quality. Therefore, in this research, a system was developed for analyzing the acoustic impulse response to determine the presence of carob moths in pomegranate fruit. For this purpose, acoustic signals generated from 100 pomegranate fruits were recorded (50 healthy and 50 infected). In order to record and extract the acoustic signals' characteristics (non-destructive test), a plastic pendulum hit the samples at three speeds (0.3, 0.9, and 1.5 m/s) and a microphone was located at three positions concerning impactor (0, 90, and 180°). For data classification, Fisher's linear discriminant classifier combined with three different feature selection methods was used. According to the statistical analysis, a speed of 0.9 m/s was selected as the optimum impact velocity. According to the classification results, the best features were natural frequency, spectral entropy, and zero-crossing, which had the least classification error. The application of these three features resulted in a classification accuracy of 97.55%.

Random projections as regularizers: learning a linear discriminant from fewer observations than dimensions

We prove theoretical guarantees for an averaging-ensemble of randomly projected Fisher linear discriminant classifiers, focusing on the case when there are fewer training observations than data dimensions. The specific form and simplicity of this ensemble permits a direct and much more detailed analysis than existing generic tools in previous works. In particular, we are able to derive the exact form of the generalization error of our ensemble, conditional on the training set, and based on this we give theoretical guarantees which directly link the performance of the ensemble to that of the corresponding linear discriminant learned in the full data space. To the best of our knowledge these are the first theoretical results to prove such an explicit link for any classifier and classifier ensemble pair. Furthermore we show that the randomly projected ensemble is equivalent to implementing a sophisticated regularization scheme to the linear discriminant learned in the original data space and this prevents overfitting in conditions of small sample size where pseudo-inverse FLD learned in the data space is provably poor. Our ensemble is learned from a set of randomly projected representations of the original high dimensional data and therefore for this approach data can be collected, stored and processed in such a compressed form. We confirm our theoretical findings with experiments, and demonstrate the utility of our approach on several datasets from the bioinformatics domain and one very high dimensional dataset from the drug discovery domain, both settings in which fewer observations than dimensions are the norm.

Block-based image steganalysis: Algorithm and performance evaluation

Traditional image steganalysis is conducted with respect to the entire image frame. In this work, we differentiate a stego image from its cover image based on steganalysis of decomposed image blocks. After image decomposition into smaller blocks, we classify image blocks into multiple classes and find a classifier for each class. Then, steganalysis of the whole image can be obtained by integrating results of all image blocks via decision fusion. Extensive performance evaluation of block-based image steganalysis is conducted. For a given test image, there exists a trade-off between the block size and the block number. We propose to use overlapping blocks to improve the steganalysis performance. Additional performance improvement can be achieved using different decision fusion schemes and different classifiers. Besides the block-decomposition framework, we point out that the choice of a proper classifier plays an important role in improving detection accuracy, and show that both the logistic classifier and the Fisher linear discriminant classifier outperforms the linear Bayes classifier by a significant margin.

Stress Emotion Recognition Based on RSP and EMG Signals

To recognize the stress emotion, a subject was put alternately in periods of high and low stress by configuring the speed and difficulty of a game named Tetris. The respiration (RSP) signal and the electromyogram (EMG) signal with different stress level were then acquired. After preprocessing, the mathematical features were calculated and automatic detection of stress level based on Fisher linear discriminant classifier was realized. The results show that the average correct detection rate of stress level based on the EMG signal can reach 97.8%. That of the RSP signal is only 86.7%. The EMG signal is more effective than the RSP signal in detection of stress level. Union of multiple physiological signals can effectively improve the correct detection rate.

Recognition Study of TCM pulse based on different Classification Methods

Pulse diagnosis is one of important diagnosis methods in Traditional Chinese Medicine (TCM). Recognition of TCM pulse has received more and more attention in recent years. Extracting proper features is crucial for satisfactory classification. While most of previous methods for feature extraction of TCM pulse have no specific correlation with the mechanism of TCM pulse, a hemodynamics method is used to calculate the pulse waveform velocity (PWV) and pulse reflection factor(Rf), which reflects the principle of TCM pulse diagnosis. Then the Fisher linear discriminant (FLD) and the K-Nearest Neighbor (KNN) algorithm are employed to classify the data and double cross-validation method is used for accuracy assessment. The average recognition rate was 80.10% by using FLD classifier and it was 87.13% by using KNN classifier. It is concluded that the PWV and Rf may be used as the features for the classification of TCM pulses.

A new computationally efficient CAD system for pulmonary nodule detection in CT imagery

Early detection of lung nodules is extremely important for the diagnosis and clinical management of lung cancer. In this paper, a novel computer aided detection (CAD) system for the detection of pulmonary nodules in thoracic computed tomography (CT) imagery is presented. The paper describes the architecture of the CAD system and assesses its performance on a publicly available database to serve as a benchmark for future research efforts. Training and tuning of all modules in our CAD system is done using a separate and independent dataset provided courtesy of the University of Texas Medical Branch (UTMB). The publicly available testing dataset is that created by the Lung Image Database Consortium (LIDC). The LIDC data used here is comprised of 84 CT scans containing 143 nodules ranging from 3 to 30mm in effective size that are manually segmented at least by one of the four radiologists. The CAD system uses a fully automated lung segmentation algorithm to define the boundaries of the lung regions. It combines intensity thresholding with morphological processing to detect and segment nodule candidates simultaneously. A set of 245 features is computed for each segmented nodule candidate. A sequential forward selection process is used to determine the optimum subset of features for two distinct classifiers, a Fisher Linear Discriminant (FLD) classifier and a quadratic classifier. A performance comparison between the two classifiers is presented, and based on this, the FLD classifier is selected for the CAD system. With an average of 517.5 nodule candidates per case/scan (517.5+/-72.9), the proposed front-end detector/segmentor is able to detect 92.8% of all the nodules in the LIDC/testing dataset (based on merged ground truth). The mean overlap between the nodule regions delineated by three or more radiologists and the ones segmented by the proposed segmentation algorithm is approximately 63%. Overall, with a specificity of 3 false positives (FPs) per case/patient on average, the CAD system is able to correctly identify 80.4% of the nodules (115/143) using 40 selected features. A 7-fold cross-validation performance analysis using the LIDC database only shows CAD sensitivity of 82.66% with an average of 3 FPs per CT scan/case.

Automatic classification of background EEG activity in healthy and sick neonates

The overall aim of our research is to develop methods for a monitoring system to be used at neonatal intensive care units. When monitoring a baby, a range of different types of background activity needs to be considered. In this work, we have developed a scheme for automatic classification of background EEG activity in newborn babies. EEG from six full-term babies who were displaying a burst suppression pattern while suffering from the after-effects of asphyxia during birth was included along with EEG from 20 full-term healthy newborn babies. The signals from the healthy babies were divided into four behavioural states: active awake, quiet awake, active sleep and quiet sleep. By using a number of features extracted from the EEG together with Fisher's linear discriminant classifier we have managed to achieve 100% correct classification when separating burst suppression EEG from all four healthy EEG types and 93% true positive classification when separating quiet sleep from the other types. The other three sleep stages could not be classified. When the pathological burst suppression pattern was detected, the analysis was taken one step further and the signal was segmented into burst and suppression, allowing clinically relevant parameters such as suppression length and burst suppression ratio to be calculated. The segmentation of the burst suppression EEG works well, with a probability of error around 4%.

Hierarchical Ensemble of Global and Local Classifiers for Face Recognition

In the literature of psychophysics and neurophysiology, many studies have shown that both global and local features are crucial for face representation and recognition. This paper proposes a novel face recognition method which exploits both global and local discriminative features. In this method, global features are extracted from the whole face images by keeping the low-frequency coefficients of Fourier transform, which we believe encodes the holistic facial information, such as facial contour. For local feature extraction, Gabor wavelets are exploited considering their biological relevance. After that, Fisher's linear discriminant (FLD) is separately applied to the global Fourier features and each local patch of Gabor features. Thus, multiple FLD classifiers are obtained, each embodying different facial evidences for face recognition. Finally, all these classifiers are combined to form a hierarchical ensemble classifier. We evaluate the proposed method using two large-scale face databases: FERET and FRGC version 2.0. Experiments show that the results of our method are impressively better than the best known results with the same evaluation protocol.

Steady-state movement related potentials for brain-computer interfacing.

An approach for brain-computer interfacing (BCI) by analysis of steady-state movement related potentials (ssMRPs) produced during rhythmic finger movements is proposed in this paper. The neurological background of ssMRPs is briefly reviewed. Averaged ssMRPs represent the development of a lateralized rhythmic potential, and the energy of the EEG signals at the finger tapping frequency can be used for single-trial ssMRP classification. The proposed ssMRP-based BCI approach is tested using the classic Fisher's linear discriminant classifier. Moreover, the influence of the current source density transform on the performance of BCI system is investigated. The averaged correct classification rates (CCRs) as well as averaged information transfer rates (ITRs) for different sliding time windows are reported. Reliable single-trial classification rates of 88%-100% accuracy are achievable at relatively high ITRs. Furthermore, we have been able to achieve CCRs of up to 93% in classification of the ssMRPs recorded during imagined rhythmic finger movements. The merit of this approach is in the application of rhythmic cues for BCI, the relatively simple recording setup, and straightforward computations that make the real-time implementations plausible.

Automatic analysis and classification of surface electromyography

In this paper, parametric modeling of surface electromyography (EMG) algorithms that facilitates automatic SEMG feature extraction and artificial neural networks (ANN) are combined for providing an integrated system for the automatic analysis and diagnosis of myopathic disorders. Three paradigms of ANN were investigated: the multilayer backpropagation algorithm, the self-organizing feature map algorithm and a probabilistic neural network model. The performance of the three classifiers was compared with that of the old Fisher linear discriminant (FLD) classifiers. The results have shown that the three ANN models give higher performance. The percentage of correct classification reaches 90%. Poorer diagnostic performance was obtained from the FLD classifier. The system presented here indicates that surface EMG, when properly processed, can be used to provide the physician with a diagnostic assist device.