Search

This paper presents a novel face detection method, which integrates the discriminating feature analysis of the input image, the statistical modeling of face and nonface classes, and the Bayes classifier for multiple frontal face detection. First, feature analysis derives a discriminating feature vector by combining the input image, its 1-D Haar wavelet representation, and its amplitude projections. Second, statistical modeling estimates the conditional probability density functions, or PDFs, of the face and nonface classes, respectively. Finally, the Bayes classifier applies the estimated conditional PDFs to detect multiple frontal faces in an image. Experimental results using 853 images (containing a total of 970 faces) from diverse image sources show the feasibility of the proposed face detection method.

Designing the optimal Bayes classifier for automated cell pattern recognition faces two major difficulties: (1) modeling and learning the conditional probabilities P(cell features--cell type) (2) developing staining independent strategies to handle staining dependent cell features while learning those conditional probabilities. In this paper, we will show such modeling and learning techniques as well as staining independent strategies. The result of the strategies tested on an automated system designed for cervical smear screening will also be reported.© (1993) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

An infrared dim and small tracking is proposed based on an explicit image filter - guided filter. The guided filter utilizes the structure in the guidance image and performs as an edge-preserving smoothing operator. The superior performance depending on the guidance image is critical advantage for target tracking. First, the guided filter can help to preserve the detail of the valuable templates and make the inaccurate ones blurry so that the tracker can distinguish the target from numerous bad templates easily. Besides, the filter can recover the content of the small target being influenced according to the guidance image, helping to alleviate the drifting problem effectively. Finally, the candidate samples are utilized to train an effective Bayes classifier to generate a robust tracker, which is easy to be implemented. Experimental results demonstrate that the presented method can track the target effectively, compared with several classical methods. Experimental results show that the proposed algorithm outperforms relative trackers in the accuracy and the robustness.

The novel smile recognition algorithm is presented based on extraction of 68 facial salient points (fp68) using the ensemble of regression trees. The smile detector exploits the Support Vector Machine linear model. It is trained with few hundreds exemplar images by SVM algorithm working in 136 dimensional space. It is shown by the strict statistical data analysis that such geometric detector strongly depends on the geometry of mouth opening area, measured by triangulation of outer lip contour. To this goal two Bayesian detectors were developed and compared with SVM detector. The first uses the mouth area in 2D image, while the second refers to the mouth area in 3D animated face model. The 3D modeling is based on Candide-3 model and it is performed in real time along with three smile detectors and statistics estimators. The mouth area/Bayesian detectors exhibit high correlation with fp68/SVM detector in a range [0:8; 1:0], depending mainly on light conditions and individual features with advantage of 3D technique, especially in hard light conditions.

An augmented set of multifractal parameters with physical interpretations have been proposed to quantify the varying distribution and shape of the multifractal spectrum. The statistical classifier with accuracy of 84.17% validates the adequacy of multi-feature MFDFA characterization of elastic scattering spectroscopy for optical diagnosis of cancer.

The United States increasingly relies on cyber-physical systems to conduct military and commercial operations. Attacks on these systems have increased dramatically around the globe. The attackers constantly change their methods, making state-of-the-art commercial and military intrusion detection systems ineffective. In this paper, we present a model to identify functional behavior of network devices from netflow traces. Our model includes two innovations. First, we define novel features for a host IP using detection of application graph patterns in IP’s host graph constructed from 5-min aggregated packet flows. Second, we present the first application, to the best of our knowledge, of Graph Semi-Supervised Learning (GSSL) to the space of IP behavior classification. Using a cyber-attack dataset collected from NetFlow packet traces, we show that GSSL trained with only 20% of the data achieves higher attack detection rates than Support Vector Machines (SVM) and Naive Bayes (NB) classifiers trained with 80% of data points. We also show how to improve detection quality by filtering out web browsing data, and conclude with discussion of future research directions.

The development of a wide area, bioaerosol early warning capability employing existing uncooled thermal imaging systems used for persistent perimeter surveillance is discussed. The capability exploits thermal imagers with other available data streams including meteorological data and employs a recursive Bayesian classifier to detect, track, and classify observed thermal objects with attributes consistent with a bioaerosol plume. Target detection is achieved based on similarity to a phenomenological model which predicts the scene-dependent thermal signature of bioaerosol plumes. Change detection in thermal sensor data is combined with local meteorological data to locate targets with the appropriate thermal characteristics. Target motion is tracked utilizing a Kalman filter and nearly constant velocity motion model for cloud state estimation. Track management is performed using a logic-based upkeep system, and data association is accomplished using a combinatorial optimization technique. Bioaerosol threat classification is determined using a recursive Bayesian classifier to quantify the threat probability of each tracked object. The classifier can accept additional inputs from visible imagers, acoustic sensors, and point biological sensors to improve classification confidence. This capability was successfully demonstrated for bioaerosol simulant releases during field testing at Dugway Proving Grounds. Standoff detection at a range of 700m was achieved for as little as 500g of anthrax simulant. Developmental test results will be reviewed for a range of simulant releases, and future development and transition plans for the bioaerosol early warning platform will be discussed.

This paper presents a comparative study of different classification algorithms for the classification of various types of inter-ply delaminations in smart composite laminates. Improved layerwise theory is used to model delamination at different interfaces along the thickness and longitudinal directions of the smart composite laminate. The input-output data obtained through surface bonded piezoelectric sensor and actuator is analyzed by the system identification algorithm to get the system parameters. The identified parameters for the healthy and delaminated structure are supplied as input data to the classification algorithms. The classification algorithms considered in this study are ZeroR, Classification via regression, Naive Bayes, Multilayer Perceptron, Sequential Minimal Optimization, Multiclass-Classifier, and Decision tree (J48). The open source software of Waikato Environment for Knowledge Analysis (WEKA) is used to evaluate the classification performance of the classifiers mentioned above via 75-25 holdout and leave-one-sample-out cross-validation regarding classification accuracy, precision, recall, kappa statistic and ROC Area.

Addressing the problems of infrared small target tracking in forward looking infrared (FLIR) system, a new infrared small target tracking method is presented, in which features binding of both target gray intensity and spatial relationship is implemented by compressive sensing so as to construct the Gaussian mixture model of compressive appearance distribution. Subsequently, naive Bayesian classification is carried out over testing samples acquired with non-uniform sampling probability to identify the most credible location of targets from background scene. A series of experiments are carried out over four infrared small target image sequences with more than 200 images for each sequence, the results demonstrate the effectiveness and advantages of the proposed method in both success rate and precision rate.

This work is devoted to the problem of the best hyperspectral images classification algorithm selection. The following algorithms are used for comparison: decision tree using full cross-validation; decision tree C 4.5; Bayesian classifier; maximum-likelihood method; MSE minimization classifier, including a special case – classification by conjugation; spectral angle classifier (for empirical mean and nearest neighbor), spectral mismatch classifier and support vector machine (SVM). There are used AVIRIS and SpecTIR hyperspectral images to conduct experiments.

Bayesian Networks have emerged in recent years as a powerful data mining technique for handling uncertainty in complex domains. The Bayesian Network represents the joint probability distribution and domain (or expert) knowledge in a compact way and provides a comprehensive method of representing relationships and influences among nodes (variables) with a graphical diagram. Actually, however, in the classification domain it was not paid attention to by researchers until the simplest of form of Bayesian Networks, Naive Bayes Classifier, turned up. Naive Bayes Classifier is a simple and efficient probability classification method, and has shown surprising performance in some domains, which owes to the independence assumption that makes Naive Bayes Classifier fit the classification more easily. However, the independence assumption obviously does not hold in the real world. Therefore, in order to meet the "naive" (unreal) assumption, this paper proposes a new image texture classification method of aerial images, PCA-NBC, which combines the Principal Components Analysis (PCA) and Naive Bayes Classifier (NBC). The PCA transforms the highly correlated features into statistically independent and orthogonal "features", so it is suitable to solve that problem and can lay a solid theoretic foundation in the application. One hundred and thirteen aerial images are used to evaluate the classification performance in the experiment. The experimental results demonstrate that the proposed method can cut down the number of features and computational costs and improve the accuracy during classification. In one word, the new method, PCA-NBC, is an attractive and effective method, which outperforms the Naive Bayes Classifier.

Feature-based methods are widely used in the brain tumor recognition system. Robust of early cancer detection is one of the most powerful image processing tools. Specifically, statistical features, such as geometric mean, harmonic mean, mean excluding outliers, median, percentiles, skewness and kurtosis, have been extracted from brain tumor glioma to aid in discriminating two levels namely, Level I and Level II using fluid attenuated inversion recovery (FLAIR) sequence in the diagnosis of brain tumor. Statistical feature describes the major characteristics of each level from glioma which is an important step to evaluate heterogeneity of cancer area pixels. In this paper, we address the task of feature selection to identify the relevant subset of features in the statistical domain, while discarding those that are either redundant or confusing, thereby improving the performance of feature-based scheme to distinguish between Level I and Level II. We apply a Decision Structure algorithm to find the optimal combination of nonhomogeneity based statistical features for the problem at hand. We employ a Naive Bayes classifier to evaluate the performance of the optimal statistical feature based scheme in terms of its glioma Level I and Level II discrimination capability and use real-data collected from 17 patients have a glioblastoma multiforme (GBM). Dataset provided from 3 Tesla MR imaging system by MD Anderson Cancer Center. For the specific data analyzed, it is shown that the identified dominant features yield higher classification accuracy, with lower number of false alarms and missed detections, compared to the full statistical based feature set. This work has been proposed and analyzed specific GBM types which Level I and Level II and the dominant features were considered as feature aid to prognostic indicators. These features were selected automatically to be better able to determine prognosis from classical imaging studies.

Sentiment analysis has become a flourishing field of text mining and natural language processing. Sentiment analysis aims to determine whether the text is written to express positive, negative, or neutral emotions about a certain domain. Most sentiment analysis researchers focus on English texts, with very limited resources available for other complex languages, such as Arabic. In this study, the target was to develop an initial model that performs satisfactorily and measures Arabic Twitter sentiment by using machine learning approach, Naive Bayes and Decision Tree for classification algorithms. The datasets used contains more than 2,000 Arabic tweets collected from Twitter. We performed several experiments to check the performance of the two algorithms classifiers using different combinations of text-processing functions. We found that available facilities for Arabic text processing need to be made from scratch or improved to develop accurate classifiers. The small functionalities developed by us in a Python language environment helped improve the results and proved that sentiment analysis in the Arabic domain needs lot of work on the lexicon side.

Healthcare associated infections (HAI) are a major threat to patient safety and are costly to health systems. Our goal is to predict the HAI performance of a hospital using the patients' experience responses as input. We use four classifiers, viz. random forest, naive Bayes, artificial feedforward neural networks, and the support vector machine, to perform the prediction of six types of HAI. The six types include blood stream, urinary tract, surgical site, and intestinal infections. Experiments show that the random forest and support vector machine perform well across the six types of HAI.

Collecting and updating forest inventory data play an important part in the forest management. The data can be obtained directly by using exact enough but low efficient ground based methods as well as from the remote sensing measurements. We present applications of airborne hyperspectral remote sensing for the retrieval of such important inventory parameters as the forest species and age composition. The hyperspectral images of the test region were obtained from the airplane equipped by the produced in Russia light-weight airborne video-spectrometer of visible and near infrared spectral range and high resolution photo-camera on the same gyro-stabilized platform. The quality of the thematic processing depends on many factors such as the atmospheric conditions, characteristics of measuring instruments, corrections and preprocessing methods, etc. An important role plays the construction of the classifier together with methods of the reduction of the feature space. The performance of different spectral classification methods is analyzed for the problem of hyperspectral remote sensing of soil and vegetation. For the reduction of the feature space we used the earlier proposed stable feature selection method. The results of the classification of hyperspectral airborne images by using the Multiclass Support Vector Machine method with Gaussian kernel and the parametric Bayesian classifier based on the Gaussian mixture model and their comparative analysis are demonstrated.

As advances in medical imaging technology are resulting in significant growth of biomedical image data, new techniques are needed to automate the process of identifying images of low quality. Automation is needed because it is very time consuming for a domain expert such as a medical practitioner or a biologist to manually separate good images from bad ones. While there are plenty of de-noising algorithms in the literature, their focus is on designing filters which are necessary but not sufficient for determining how useful an image is to a domain expert. Thus a computational tool is needed to assign a score to each image based on its perceived quality. In this paper, we introduce a machine learning-based score and call it the Quality of Image (QoI) score. The QoI score is computed by combining the confidence values of two popular classification techniques—support vector machines (SVMs) and Naive Bayes classifiers. We test our technique on clinical image data obtained from cancerous tissue samples. We used 747 tissue samples that are stained by four different markers (abbreviated as CK15, pck26, E_cad and Vimentin) leading to a total of 2,988 images. The results show that images can be classified as good (high QoI), bad (low QoI) or ugly (intermediate QoI) based on their QoI scores. Our automated labeling is in agreement with the domain experts with a bi-modal classification accuracy of 94%, on average. Furthermore, ugly images can be recovered and forwarded for further post-processing.

In this paper, a new neural network directed Bayes decision rule is developed for target classification exploiting the target's dynamic behavior. The system consists of a feature extractor, a neural network directed conditional probability generator and a sequential Bayes classifier. The velocity and curvature sequences extracted from each track are used as the primary features. Several hidden states are used to train the neural network, the output of which is the conditional probability of occurring the hidden states given the observations. These conditional probabilities are then used as the inputs to the Bayes classifier to make the classification. The classification results are updated recursively whenever a new scan of data is received. Simulation results on both clean tracks and heavily cluttered Infrared (IR) satellite images are presented to demonstrate the effectiveness of the proposed methods.

The important step of data preprocessing of data mining is feature selection. Feature selection is used to improve the performance of data mining algorithms by removing the irrelevant and redundant features. By positional cross-identification, the multi-wavelength data of 1656 active galactic nuclei (AGNs), 3718 stars, and 173 galaxies are obtained from optical (USNO-A2.0), X-ray (ROSAT), and infrared (Two Micron All- Sky Survey) bands. In this paper we applied a kind of filter approach named ReliefF to select features from the multi-wavelength data. Then we put forward the naive Bayes classifier to classify the objects with the feature subsets and compare the results with and without feature selection, and those with and without adding weights to features. The result shows that the naive Bayes classifier based on ReliefF algorithms is robust and efficient to preselect AGN candidates.

The paper presents improvement of the method of environmental monitoring of water bodies based on bioindication by phytoplankton, which identify phytoplankton particles carried out on the basis of comparison array multispectral images using Bayesian classifier of solving function based on Mahalanobis distance. It allows to evaluate objectively complex anthropogenic and technological impacts on aquatic ecosystems.

In this paper, we present a hierarchical kernel associative memory (H-KAM) based computational model with Finite Ridgelet Transform (FRIT) representation for multispectral palmprint recognition. To characterize a multispectral palmprint image, the Finite Ridgelet Transform is used to achieve a very compact and distinctive representation of linear singularities while it also captures the singularities along lines and edges. The proposed system makes use of Finite Ridgelet Transform to represent multispectral palmprint image and it is then modeled by Kernel Associative Memories. Finally, the recognition scheme is thoroughly tested with a benchmarking multispectral palmprint database CASIA. For recognition purpose a Bayesian classifier is used. The experimental results exhibit robustness of the proposed system under different wavelengths of palm image.

The target tracking by the spatio-temporal learning is a kind of online tracking algorithm based on Bayesian framework. But it has the excursion problem when applied in the infrared dim target. Based on the principle of the spatio-temporal learning algorithm, the excursion problem was analyzed and a new robust algorithm for infrared dim target tracking is proposed in this paper. Firstly, the Guide Image Filter was adopted to process the input image to preserve edges and eliminate the noise of the image. Secondly, the ideal spatial context model was calculated with the input image that contains little noise, which can be got by subtracting the filtering result from the original image. Simultaneously, a new weight in the context prior model was proposed to indicate that the prior is also related to the local gray level difference. The performance of the presented algorithm was tested with two infrared air image sequences, and the experimental results show that the proposed algorithm performs well in terms of efficiency, accuracy and robustness.

Stroke is the third most common cause of death in developed countries. Clinical trials are currently investigating whether advanced Computed Tomography can be of benefit for diagnosing stroke at the acute phase. These trials are based on large patients cohorts that need to be manually annotated to obtain a reference standard of tissue loss at follow-up, resulting in extensive workload for the radiologists. Therefore, there is a demand for accurate and reliable automatic lesion segmentation methods. This paper presents a novel method for the automatic detection and segmentation of ischemic lesions in CT images. The method consists of multiple sequential stages. In the initial stage, pixel classification is performed using a naive Bayes classifier in combination with a tissue homogeneity algorithm in order to localize ischemic lesion candidates. In the next stage, the candidates are segmented using a marching cubes algorithm. Regional statistical analysis is used to extract features based on local information as well as contextual information from the contra-lateral hemisphere. Finally, the extracted features are summarized into a likelihood of ischemia by a supervised classifier. An area under the Receiver Operating Characteristic curve of 0.91 was obtained for the identification of ischemic lesions. The method performance on lesion segmentation reached a Dice similarity coeficient (DSC) of 0.74±0.09, whereas an independent human observer obtained a DSC of 0.79±0.11 in the same dataset. The experiments showed that it is feasible to automatically detect and segment ischemic lesions in CT images, obtaining a comparable performance as human observers.

With the development of Information Technology , people have entered the era of Big Data, and the demand for intelligent information is more intense. How to make computer provide more personalized and efficient service for all walks of life, is something worth exploring. In this paper, we aim to predict user’s character by analyzing the textual content of his/her micro-blog, which is the foundation of Personalized Service . Our study describes the method of creating a prediction model about user’s character by using Bayesian algorithms. Experimental results show that the Naive Bayes approach is a valid and promoted analytic method in micro-blog character analysis.

Content Based Image Retrieval (CBIR) system could exploit the wealth of High-Resolution Computed Tomography (HRCT) data stored in the archive by finding similar images to assist radiologists for self learning and differential diagnosis of Interstitial Lung Diseases (ILDs). HRCT findings of ILDs are classified into several categories (e.g. consolidation, emphysema, ground glass, nodular etc.) based on their texture like appearances. Therefore, analysis of ILDs is considered as a texture analysis problem. Many approaches have been proposed for CBIR of lung images using texture as primitive visual content. This paper presents a new approach to CBIR for ILDs. The proposed approach makes use of a trained neural network (NN) to find the output class label of query image. The degree of confidence of the NN classifier is analyzed using Naive Bayes classifier that dynamically takes a decision on the size of the search space to be used for retrieval. The proposed approach is compared with three simple distance based and one classifier based texture retrieval approaches. Experimental results show that the proposed technique achieved highest average percentage precision of 92.60% with lowest standard deviation of 20.82%.

Due to the promotion of lung cancer screening, more Stage I non-small-cell lung cancers (NSCLC) are currently detected, which usually have favorable prognosis. However, a high percentage of the patients have cancer recurrence after surgery, which reduces overall survival rate. To achieve optimal efficacy of treating and managing Stage I NSCLC patients, it is important to develop more accurate and reliable biomarkers or tools to predict cancer prognosis. The purpose of this study is to investigate a new quantitative image analysis method to predict the risk of lung cancer recurrence of Stage I NSCLC patients after the lung cancer surgery using the conventional chest computed tomography (CT) images and compare the prediction result with a popular genetic biomarker namely, protein expression of the excision repair cross-complementing 1 (ERCC1) genes. In this study, we developed and tested a new computer-aided detection (CAD) scheme to segment lung tumors and initially compute 35 tumor-related morphologic and texture features from CT images. By applying a machine learning based feature selection method, we identified a set of 8 effective and non-redundant image features. Using these features we trained a naive Bayesian network based classifier to predict the risk of cancer recurrence. When applying to a test dataset with 79 Stage I NSCLC cases, the computed areas under ROC curves were 0.77±0.06 and 0.63±0.07 when using the quantitative image based classifier and ERCC1, respectively. The study results demonstrated the feasibility of improving accuracy of predicting cancer prognosis or recurrence risk using a CAD-based quantitative image analysis method.

The constrained energy minimization (CEM) algorithm and the closely related matched filter processor have been widely used for target detection in hyperspectral data exploitation applications. In this paper, we look at the key assumptions underlying the derivation of each algorithm and the effect these assumptions have upon their performance. To illuminate and better understand their operation, we compare both algorithms to Fisher's linear discriminant and the quadratic Bayes classifier. Bayes classifier reduces to Fisher's linear discriminant when the target and background covariances are equal. Furthermore, Fisher's linear discriminant is reduced to the matched filter, when we look for low probability targets. These interrelations can be used to justify the use of the matched filter, which has been developed for the detection of known signals in additive noise, for hyperspectral target which are not corrupted by additive noise. Finally, we investigate under what conditions the output of the matched filter follows a normal distribution.

This paper is concerned with the identification of grassland management using both optical and radar data. In that context, grazing, mowing and a mix of these two managements are commonly used by the farmers on grassland fields. These practices and their intensity of use have different environmental impact. Thus, the objectives of this study are, firstly, to identify grassland management practices using a time series of optical and radar imagery at high spatial resolution and, secondly, to evaluate the contribution of radar data to improve identification of farming practices on grasslands. Because of cloud coverage and revisit frequency of satellite, the number of available optical data is limited during the vegetation period. Thus, radar data can be considered as an ideal complement. The present study is based on the use of SPOT, Landsat and RADARSAT-2 data, acquired in 2010 during the growing period. After a pre-processing step, several vegetation indices, biophysical variables, backscattering coefficients and polarimetric discriminators were computed on the data set. Then, with the help of some statistics, the most discriminating variables have been identified and used to classify grassland fields. In addition, to take into account the temporal variation of variables, dedicated indexes as first and second order derivatives were used. Classification process was based on training samples resulting from field campaigns and computed according six methods: Decision Trees, K-Nearest Neighbor, Neural Networks, Support Vector Machines, the Naive Bayes Classifier and Linear Discriminant Analysis. Results show that combined use of optical and radar remote sensing data is not more efficient for grassland management identification.

In this paper, a Synthetic Aperture Radar Automatic Target Recognition approach based on Gaussian process (GP) classification is proposed. It adopts kernel principal component analysis to extract sample features and implements target recognition by using GP classification with automatic relevance determination (ARD) function. Compared with k-Nearest Neighbor, Naive Bayes classifier and Support Vector Machine, GP with ARD has the advantage of automatic model selection and hyper-parameter optimization. The experiments on UCI datasets and MSTAR database show that our algorithm is self-tuning and has better recognition accuracy as well.

A process for selecting a sparse subset of features that maximize discrimination between target classes is described in a Bayesian framework. Demonstrated on high range resolution radar (HRR) signature data, this has the effect of selecting the most informative range bins for a classification task. The sparse Bayesian classifier (SBC) model is directly compared against Fisher's linear discriminant analysis (LDA), showing a clear performance gain with the Bayesian framework using HRRs from the publicly available MSTAR data set. The discriminative power of the selected features from the SBC is shown to be particularly dominant over LDA when only a few features are selected or when there is a shift in training and testing data sets, as demonstrated by training on a specific target type and testing on a slightly different target type.

The literature is abundant with papers on gender classificati on research. However the majority of such research is based on the assumption that there is enough resolution so that the subject’s face can be resolved . Hence the majority of the research is actually in the face recognition and facial feature area. A gap exists for gender classification under challenging operating conditions—different seasonal conditions, different clothing, etc.—and when the subject’s face cannot be resolved due to lack of reso lution. The Seasonal Weather and Gender (SWAG) Database is a novel database that contains subjects walking through a scene under operating conditions that span a calendar year. This paper exploits a subset of that database—the SWAG One dataset—using data mining techniques, traditional classifiers (ex. Naive Bayes, Support Vector Machine, etc.) and traditional (canny edge det ection, etc.) and non-traditional (height/width ratios, etc.) feature extractors to achieve high correct gender classification rates (greater than 85%). Another novelty includes exploiting frame differentials. Keywords: Gender classification, SWAG One dataset, robust operating conditions, data mining

Protein crystallization plays a crucial role in pharmaceutical research by supporting the investigation of a protein’s molecular structure through X-ray diffraction of its crystal. Due to the rare occurrence of crystals, images must be manually inspected, a laborious process. We develop a solution incorporating a regularized, logistic regression model for automatically evaluating these images. Standar d image features, such as shape context, Gabor filters and Fourier transforms, are first extracted to represent the heterogeneous appearance of our images. Then the proposed solution utilizes Elastic Net to select relevant features. Its L 1 -regularization mitigates the effects of our large dataset, and its L 2 -regularization ensures proper operation when the feature number exceeds the sample number. A two -tier cascade classifier based on naive Bayes and random forest algorithm s categorized the images. In order to validate the proposed method, we experimentally compare it with naive Bayes, linear discriminant analysis, random forest, and their two -tier cascade classifiers, by 10-fold cross validation. Our experimental results demonstrate -category a 3 accuracy of 74% , outperforming other models. In addition, Elastic Net better reduces the false negatives responsible for a high, domain specific risk. To the best of our knowledge, this is the first attempt to apply Elastic Net to classifying protein crysta llization images. Performance measured a large pharmaceutical dataset also fared well in comparison with onthose presented in the previous studies, while the reduction of the high- risk false negatives is promising.

Model observers (MO) are widely used in medical imaging to act as surrogates of human observers in task-based image quality evaluation, frequently towards optimization of reconstruction algorithms. In SPECT myocardial perfusion imaging (MPI), a realistic task-based approach involves detection and localization of perfusion defects, as well as a subsequent assessment of defect severity. In this paper we explore a machine-learning MO based on Naive- Bayes classification (NB-MO). NB-MO uses a set of polar-map image features to predict lesion detection, localization and severity scores given by five human readers for a set of simulated 3D SPECT-MPI patients. The simulated dataset included lesions with different sizes, perfusion-reduction ratios, and locations. Simulated projections were reconstructed using two readily used methods namely: FBP and OSEM. For validation, a multireader multi-case (MRMC) analysis of alternative free-response ROC (AFROC) curve was performed for NB-MO and human observers. For comparison, we also report performances of a statistical Hotelling Observer applied on polar-map images. Results show excellent agreement between NB-MO and humans, as well as model’s good generalization between different reconstruction treatments.

This application paper presents a machine vision solution for the quality inspection of flat glass products. A contact image sensor (CIS) is used to generate digital images of the glass surfaces. The presented machine vision based quality inspection at the end of the production line aims to classify five different glass defect types. The defect images are usually characterized by very little ‘image structure’, i.e. homogeneous regions without distinct image texture. Additionally, these defect images usually consist of only a few pixels. At the same time the appearance of certain defect classes can be very diverse (e.g. water drops). We used simple state-of-the-art image features like histogram-based features (std. deviation, curtosis, skewness), geometric features (form factor/elongation, eccentricity, Hu-moments) and texture features (grey level run length matrix, co-occurrence matrix) to extract defect information. The main contribution of this work now lies in the systematic evaluation of various machine learning algorithms to identify appropriate classification approaches for this specific class of images. In this way, the following machine learning algorithms were compared: decision tree (J48), random forest, JRip rules, naive Bayes, Support Vector Machine (multi class), neural network (multilayer perceptron) and k-Nearest Neighbour. We used a representative image database of 2300 defect images and applied cross validation for evaluation purposes.

Multivariate images are now commonly produced in many applications. If their process is possible due to computers power and new programming languages, theoretical difficulties have still to be solved. Standard image analysis operators are defined for scalars rather than for vectors and their extension is not immediate. Several solutions exist but their pertinence is hardly linked to context. In the present paper we are going to get interested in segmentation of vector images also including a priori knowledge. The proposed strategy combines a decision procedure (where points are classified) and an automatic segmentation scheme (where regions are properly extracted). The classification is made using a Bayesian classifier. The segmentation is computed via a region growing method: the morphological Watershed transform. A direct computation of the Watershed transform on vector images is not possible since vector sets are not ordered. So, the Bayesian classifier is used for computing a scalar distance map where regions are enhanced or attenuated depending on their similitude to a reference shape: the current distance is the Mahalanobis distance. This combination allows to transfer the decision function from pixels to regions and to preserve the advantages of the original Watershed transform defined for scalar functions. The algorithm is applied for segmenting colour images (with a priori) and medical images, especially dermatology images where skin lesions have to be detected.

Classification and characterization of text is of ever growing importance in defense and national security. The text classification task is an instance of classification using sparse features residing in a high dimensional feature space. Two standard (of a wide selection of available) algorithms for this task are the naive Bayes classifier and the Rocchio linear classifier. Naive Bayes classifiers are widely applied; the Rocchio algorithm is primarily used in document classification and information retrieval. Both these classifiers are popular because of their simplicity and ease of application, computational speed and reasonable performance. One aspect of the Rocchio approach, inherited from its information retrieval origin, is that it explicitly uses both positive and negative models. Parameters have been introduced which make it adaptive to the particulars of the corpora of interest and thereby improve its performance. The ideas inherent in these classifiers and in second generation wavelets can be recombined into new algorithms for classification. An example is a classifier using second generation wavelet-like functions for class probes that mimic the Rocchio positive template - negative template approach.

One of the key requirements of real-time processing systems for remote sensors is the ability to accurately and automatically geo-locate events. This capability often relies on the ability to find control points to feed into a registration-based geo-location algorithm. Clouds can make the choice of control points difficult. If each pixel in a given image can be identified as cloudy or clear, the geo-location algorithm can limit the control point selection to clear pixels, thereby improving registration accuracy. Most cloud masking algorithms rely on a large number of spectral bands for good results, e.g., MODIS, whereas with our sensor, we have only three simultaneous bands available. This paper discusses a promising new approach to generating cloud masks in real-time with a limited number of spectral bands. The effort investigated statistical methods, spatial and texture-based approaches and evaluated performance on real remote sensing data. Although the spatial and texture-based approaches did not exhibit good performance due to sensor limitations in spatial resolution and too much variation in spectral response of both surface features and clouds, the statistical classification approach applied to only two bands performed very well. Images from three daytime remote sensing collects were analyzed to determine features that best separate pixels into cloudy and clear classes. A Bayes classifier was then applied to feature vectors computed for each pixel to generate a binary cloud mask. Initial results are excellent and show very good accuracy over a variety of terrain types, including mountains, desert, and coastline.

Music genres are the simplest and effect descriptors for searching music libraries stores or catalogues. The paper compares the results of two automatic music genres classification systems implemented by using two different yet simple classifiers (K-Nearest Neighbor and Naive Bayes). First a 10-12 second sample is selected and features are extracted from it, and then based on those features results of both classifiers are represented in the form of accuracy table and confusion matrix. An experiment carried out on test 60 taken from middle of a song represents the true essence of its genre as compared to the samples taken from beginning and ending of a song. The novel techniques have achieved an accuracy of 91% and 78% by using Naive Bayes and KNN classifiers respectively.

The challenge of rapid footstep detection and classification in remote locations has long been an important area of study for defense technology and national security. Also, as the military seeks to create effective and disposable unattended ground sensors (UGS), computational complexity and power consumption have become essential considerations in the development of classification techniques. In response to these issues, a research project at the Flexible Display Center at Arizona State University (ASU) has experimented with footstep classification using the matching pursuit decomposition (MPD) time-frequency analysis method. The MPD provides a parsimonious signal representation by iteratively selecting matched signal components from a pre-determined dictionary. The resulting time-frequency representation of the decomposed signal provides distinctive features for different types of footsteps, including footsteps during walking or running activities. The MPD features were used in a Bayesian classification method to successfully distinguish between the different activities. The computational cost of the iterative MPD algorithm was reduced, without significant loss in performance, using a modified MPD with a dictionary consisting of signals matched to cadence temporal gait patterns obtained from real seismic measurements. The classification results were demonstrated with real data from footsteps under various conditions recorded using a low-cost seismic sensor.

Classification of human and animal targets imaged by a linear pyroelectic array senor presents some unique challenges especially in target segmentation and feature extraction. In this paper, we apply two approaches to address this problem. Both techniques start with the variational energy functional level set segmentation technique to separate the object from background. After segmentation, in the first technique, we extract features such as texture, invariant moments, edge, shape information, and spectral contents of the segmented object. These features are fed to classifiers including Naive Bayesian (NB), and Support Vector Machine (SVM) for human against animal classification. In the second technique, the speeded up robust feature (SURF) extraction algorithm is applied to the segmented objects. A code book technique is used to classify objects based on SURF features. Human and animal data acquired-using the pyroelectric sensor in different terrains, are used for performance evaluation of the algorithms. The evaluation indicates that the features extracted in the first technique in conjunction with the NB classifier provide the highest classification rates. While the SURF feature plus code book approach provides a slightly lower classification rate, it provides better computational efficiency lending itself to real time implementation.

Current development of algorithms in computer-aided diagnosis (CAD) scheme is growing rapidly to assist the radiologist in medical image interpretation. Texture analysis of computed tomography (CT) scans is one of important preliminary stage in the computerized detection system and classification for lung cancer. Among different types of images features analysis, Haralick texture with variety of statistical measures has been used widely in image texture description. The extraction of texture feature values is essential to be used by a CAD especially in classification of the normal and abnormal tissue on the cross sectional CT images. This paper aims to compare experimental results using texture extraction and different machine leaning methods in the classification normal and abnormal tissues through lung CT images. The machine learning methods involve in this assessment are Artificial Immune Recognition System (AIRS), Naive Bayes, Decision Tree (J48) and Backpropagation Neural Network. AIRS is found to provide high accuracy (99.2%) and sensitivity (98.0%) in the assessment. For experiments and testing purpose, publicly available datasets in the Reference Image Database to Evaluate Therapy Response (RIDER) are used as study cases. © (2013) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.

Persian language is the official language of Iran, Tajikistan and Afghanistan. Local online users often represent their opinions and experiences on the web with written Persian. Although the information in those reviews is valuable to potential consumers and sellers, the huge amount of web reviews make it difficult to give an unbiased evaluation to a product. In this paper, standard machine learning techniques SVM and naive Bayes are incorporated into the domain of online Persian Movie reviews to automatically classify user reviews as positive or negative and performance of these two classifiers is compared with each other in this language. The effects of feature presentations on classification performance are discussed. We find that accuracy is influenced by interaction between the classification models and the feature options. The SVM classifier achieves as well as or better accuracy than naive Bayes in Persian movie. Unigrams are proved better features than bigrams and trigrams in capturing Persian sentiment orientation.

Diabetic retinopathy (DR) affects more than 4.4 million Americans age 40 and over. Automatic screening for DR has shown to be an efficient and cost-effective way to lower the burden on the healthcare system, by triaging diabetic patients and ensuring timely care for those presenting with DR. Several supervised algorithms have been developed to detect pathologies related to DR, but little work has been done in determining the size of the training set that optimizes an algorithm’s performance. In this paper we analyze the effect of the training sample size on the performance of a top-down DR screening algorithm for different types of statistical classifiers. Results are based on partial least squares (PLS), support vector machines (SVM), k-nearest neighbor (kNN), and Naive Bayes classifiers. Our dataset consisted of digital retinal images collected from a total of 745 cases (595 controls, 150 with DR). We varied the number of normal controls in the training set, while keeping the number of DR samples constant, and repeated the procedure 10 times using randomized training sets to avoid bias. Results show increasing performance in terms of area under the ROC curve (AUC) when the number of DR subjects in the training set increased, with similar trends for each of the classifiers. Of these, PLS and k-NN had the highest average AUC. Lower standard deviation and a flattening of the AUC curve gives evidence that there is a limit to the learning ability of the classifiers and an optimal number of cases to train on.

In this paper we describe a modified classification method destined for extractive summarization purpose. The classification in this method doesn’t need a learning corpus; it uses the input text to do that. First, we cluster the document sentences to exploit the diversity of topics, then we use a learning algorithm (here we used Naive Bayes) on each cluster considering it as a class. After obtaining the classification model, we calculate the score of a sentence in each class, using a scoring model derived from classification algorithm. These scores are used, then, to reorder the sentences and extract the first ones as the output summary. We conducted some experiments using a corpus of scientific papers, and we have compared our results to another summarization system called UNIS.1 Also, we experiment the impact of clustering threshold tuning, on the resulted summary, as well as the impact of adding more features to the classifier. We found that this method is interesting, and gives good performance, and the addition of new features (which is simple using this method) can improve summary’s accuracy.

Correlation engines have been evolving since the implementation of radar. Here, correlation refers to association, which in this context is track-to-track association. In modern sensor fusion architectures, correlation and sensor registration are required to produce common, continuous, and unambiguous tracks of all objects in the surveillance area. The objective is to provide a unified picture of the theatre or area of interest to battlefield decision makers. This unified picture has many names, but is most commonly referred to as a Single Integrated Picture (SIP). A related process, known as sensor registration or gridlock filtering (gridlocking), refers to the reduction in navigation errors and sensor misalignment errors so that one sensor's track data can be accurately transformed into another sensor's coordinate system. As platforms gain multiple sensors, the correlation and gridlocking of tracks become significantly more difficult. Current correlation technology revolves around likelihood ratio theory and the assignment algorithm to resolve association ambiguities. While a Bayes classifier is the best classifier, all classifiers potentially lead to classification errors. In this paper, we examine the track association and sensor registration problem in terms of several correlation classifiers, the most famous of these being the matched filter. Thus, we seek some unification between the term correlation with regards to track association and correlation with regards to pattern recognition. We examine several classes of correlation classifiers and discuss their application to the generation of a SIP when coupled with a sensor registration algorithm. The availability of these techniques on optical processing platforms is an obvious benefit to track association. We briefly discuss the implementation of some of these techniques on a commercial frequency plane correlator.

In this paper, we investigate several fusion techniques for designing a composite classifier to improve the performance (probability of correct classification) of FLIR ATR. The motivation behind the fusion of ATR algorithms is that if each contributing technique in a fusion algorithm composite classifier) emphasizes on learning at least some features of the targets that are not learned by other contributing techniques for making a classification decision, a fusion of ATR algorithms may improve overall probability of correct classification of the composite classifier. In this research, we propose to use four ATR algorithms for fusion. We propose to use averaged Bayes classifier, committee of experts, stacked-generalization, winner-takes-all, and ranking-based fusion techniques for designing the composite classifiers. The experimental results show an improvement of more than 5 % over the best individual performance.© (2002) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

In this paper, we investigate several fusion techniques for designing a composite classifier to improve the performance (probability of correct classification) of FLIR ATR. The motivation behind the fusion of ATR algorithms is that if each contributing technique in a fusion algorithm (composite classifier) emphasizes on learning at least some features of the targets that are not learned by other contributing techniques for making a classification decision, a fusion of ATR algorithms may improve overall probability of correct classification of the composite classifier. In this research, we propose to use four ATR algorithms for fusion. We propose to use averaged Bayes classifier, committee of experts, stacked-generalization, winner-takes-all, and ranking-based fusion techniques for designing the composite classifiers. The experimental results show an improvement of more than 5 % over the best individual performance.

In our previous work, we showed that fractal-based texture features are effective in detection, segmentation and classification of posterior-fossa (PF) pediatric brain tumor in multimodality MRI. We exploited an information theoretic approach such as Kullback-Leibler Divergence (KLD) for feature selection and ranking different texture features. We further incorporated the feature selection technique with segmentation method such as Expectation Maximization (EM) for segmentation of tumor T and non tumor (NT) tissues. In this work, we extend the two class KLD technique to multiclass for effectively selecting the best features for brain tumor (T), cyst (C) and non tumor (NT). We further obtain segmentation robustness for each tissue types by computing Bay's posterior probabilities and corresponding number of pixels for each tissue segments in MRI patient images. We evaluate improved tumor segmentation robustness using different similarity metric for 5 patients in T1, T2 and FLAIR modalities.

This study compares the performance of two proven but very different machine learners, Naive Bayes and logistic regression, for differentiating malignant and benign breast masses using ultrasound imaging. Ultrasound images of 266 masses were analyzed quantitatively for shape, echogenicity, margin characteristics, and texture features. These features along with patient age, race, and mammographic BI-RADS category were used to train Naive Bayes and logistic regression classifiers to diagnose lesions as malignant or benign. ROC analysis was performed using all of the features and using only a subset that maximized information gain. Performance was determined by the area under the ROC curve, Az, obtained from leave-one-out cross validation. Naive Bayes showed significant variation (Az 0.733 ± 0.035 to 0.840 ± 0.029, P < 0.002) with the choice of features, but the performance of logistic regression was relatively unchanged under feature selection (Az 0.839 ± 0.029 to 0.859 ± 0.028, P = 0.605). Out of 34 features, a subset of 6 gave the highest information gain: brightness difference, margin sharpness, depth-to-width, mammographic BI-RADs, age, and race. The probabilities of malignancy determined by Naive Bayes and logistic regression after feature selection showed significant correlation (R 2 = 0.87, P < 0.0001). The diagnostic performance of Naive Bayes and logistic regression can be comparable, but logistic regression is more robust. Since probability of malignancy cannot be measured directly, high correlation between the probabilities derived from two basic but dissimilar models increases confidence in the predictive power of machine learning models for characterizing solid breast masses on ultrasound.

This paper proposes a palmprint identification system using Finite Ridgelet Transform (FRIT) and Bayesian classifier. FRIT is applied on the ROI (region of interest), which is extracted from palmprint image, to extract a set of distinctive features from palmprint image. These features are used to classify with the help of Bayesian classifier. The proposed system has been tested on CASIA and IIT Kanpur palmprint databases. The experimental results reveal better performance compared to all well known systems.

Motion-based artifacts lead in breast MRI to diagnostic misinterpretation and therefore represents an important prerequisite to automatic lesion detection and diagnosis. In the present paper, we evaluate the performance of a computer-aided diagnosis (CAD) system consisting of motion correction, lesion segmentation, feature extraction and classification. Many novel feature extraction techniques are proposed and tested in conjunction with motion correction and classification. Our simulation results have shown that motion compensation combined with Minkowsi functionals and Bayesian classifier can improve the lesion detection and classification.