scispace - formally typeset
Search or ask a question

Showing papers in "Pattern Analysis and Applications in 2002"


Journal ArticleDOI
TL;DR: Simulation studies show that the performance of the combining techniques is strongly affected by the small sample size properties of the base classifier: boosting is useful for large training sample sizes, while bagging and the random subspace method are useful for criticalTraining sample sizes.
Abstract: Recently bagging, boosting and the random subspace method have become popular combining techniques for improving weak classifiers. These techniques are designed for, and usually applied to, decision trees. In this paper, in contrast to a common opinion, we demonstrate that they may also be useful in linear discriminant analysis. Simulation studies, carried out for several artificial and real data sets, show that the performance of the combining techniques is strongly affected by the small sample size properties of the base classifier: boosting is useful for large training sample sizes, while bagging and the random subspace method are useful for critical training sample sizes. Finally, a table describing the possible usefulness of the combining techniques for linear classifiers is presented.

449 citations


Journal ArticleDOI
TL;DR: This review is organised into five major sections, covering a general overview, Arabic writing characteristics, Arabic text recognition system, Arabic OCR software and conclusions.
Abstract: Off-line recognition requires transferring the text under consideration into an image file. This represents the only available solution to bring the printed materials to the electronic media. However, the transferring process causes the system to lose the temporal information of that text. Other complexities that an off-line recognition system has to deal with are the lower resolution of the document and the poor binarisation, which can contribute to readability when essential features of the characters are deleted or obscured. Recognising Arabic script presents two additional challenges: orthography is cursive and letter shape is context sensitive. Certain character combinations form new ligature shapes, which are often font-dependent. Some ligatures involve vertical stacking of characters. Since not all letters connect, word boundary location becomes an interesting problem, as spacing may separate not only words, but also certain characters within a word. Various techniques have been implemented to achieve high recognition rates. These techniques have tackled different aspects of the recognition system. This review is organised into five major sections, covering a general overview, Arabic writing characteristics, Arabic text recognition system, Arabic OCR software and conclusions.

207 citations


Journal ArticleDOI
TL;DR: This paper introduces a hierarchical technique to recursively decompose a C-class problem into C_1 two-(meta) class problems, and introduces a generalised modular learning framework used to partition a set of classes into two disjoint groups called meta-classes.
Abstract: Many classification problems involve high dimensional inputs and a large number of classes. Multiclassifier fusion approaches to such difficult problems typically centre around smart feature extraction, input resampling methods, or input space partitioning to exploit modular learning. In this paper, we investigate how partitioning of the output space (i.e. the set of class labels) can be exploited in a multiclassifier fusion framework to simplify such problems and to yield better solutions. Specifically, we introduce a hierarchical technique to recursively decompose a C-class problem into C_1 two-(meta) class problems. A generalised modular learning framework is used to partition a set of classes into two disjoint groups called meta-classes. The coupled problems of finding a good partition and of searching for a linear feature extractor that best discriminates the resulting two meta-classes are solved simultaneously at each stage of the recursive algorithm. This results in a binary tree whose leaf nodes represent the original C classes. The proposed hierarchical multiclassifier framework is particularly effective for difficult classification problems involving a moderately large number of classes. The proposed method is illustrated on a problem related to classification of landcover using hyperspectral data: a 12-class AVIRIS subset with 180 bands. For this problem, the classification accuracies obtained were superior to most other techniques developed for hyperspectral classification. Moreover, the class hierarchies that were automatically discovered conformed very well with human domain experts’ opinions, which demonstrates the potential of using such a modular learning approach for discovering domain knowledge automatically from data.

207 citations


Journal ArticleDOI
Tin Kam Ho1
TL;DR: There are strong correlations between the classifier accuracies and measures of length of class boundaries, thickness of the class manifolds, and nonlinearities of decision boundaries and the bootstrapping method is better when the classes are compact and the boundaries are smooth.
Abstract: Using a number of measures for characterising the complexity of classification problems, we studied the comparative advantages of two methods for constructing decision forests – bootstrapping and random subspaces We investigated a collection of 392 two-class problems from the UCI depository, and observed that there are strong correlations between the classifier accuracies and measures of length of class boundaries, thickness of the class manifolds, and nonlinearities of decision boundaries We found characteristics of both difficult and easy cases where combination methods are no better than single classifiers Also, we observed that the bootstrapping method is better when the training samples are sparse, and the subspace method is better when the classes are compact and the boundaries are smooth

198 citations


Journal ArticleDOI
TL;DR: Experimental evidence highlights the gain in quality resulting from combining some of the most widely used prediction methods with the authors' SVMs rather than with the ensemble methods traditionally used in the field, which increases when the outputs of the combiners are post-processed with a DP algorithm.
Abstract: The idea of performing model combination, instead of model selection, has a long theoretical background in statistics. However, making use of theoretical results is ordinarily subject to the satisfaction of strong hypotheses (weak error correlation, availability of large training sets, possibility to rerun the training procedure an arbitrary number of times, etc.). In contrast, the practitioner is frequently faced with the problem of combining a given set of pre-trained classifiers, with highly correlated errors, using only a small training sample. Overfitting is then the main risk, which cannot be overcome but with a strict complexity control of the combiner selected. This suggests that SVMs should be well suited for these difficult situations. Investigating this idea, we introduce a family of multi-class SVMs and assess them as ensemble methods on a real-world problem. This task, protein secondary structure prediction, is an open problem in biocomputing for which model combination appears to be an issue of central importance. Experimental evidence highlights the gain in quality resulting from combining some of the most widely used prediction methods with our SVMs rather than with the ensemble methods traditionally used in the field. The gain increases when the outputs of the combiners are post-processed with a DP algorithm.

127 citations


Journal ArticleDOI
TL;DR: This paper presents four state-of-the-art systems in the third class, based on the fusion of boundary/surface-based with region-based techniques outlined in Part II of the paper, also called regional-geometric deformation models, which take the paradigm of partial differential equations in the level set framework.
Abstract: Extensive growth in functional brain imaging, perfusion-weighted imaging, diffusion-weighted imaging, brain mapping and brain scanning techniques has led tremendously to the importance of the cerebral cortical segmentation, both in 2-D and 3-D, from volumetric brain magnetic resonance imaging data sets Besides that, recent growth in deformable brain segmentation techniques in 2-D and 3-D has brought the engineering community, such as the areas of computer vision, image processing, pattern recognition and graphics, closer to the medical community, such as to neuro-surgeons, psychiatrists, oncologists, neuro-radiologists and internists This paper is an attempt to review the state-of-the-art 2-D and 3-D cerebral cortical segmentation techniques from brain magnetic resonance imaging based on three main classes: region-based, boundary/surface-based and fusion of boundary/surface-based with region-based techniques In the first class, region-based techniques, we demonstrated more than 18 different techniques for segmenting the cerebral cortex from brain slices acquired in orthogonal directions In the second class, boundary/surface-based, we showed more than ten different techniques to segment the cerebral cortex from magnetic resonance brain volumes Particular emphasis will be placed by presenting four state-of-the-art systems in the third class, based on the fusion of boundary/surface-based with region-based techniques outlined in Part II of the paper, also called regional-geometric deformation models, which take the paradigm of partial differential equations in the level set framework We also discuss the pros and cons of various techniques, besides giving the mathematical foundations for each sub-class in the cortical taxonomy

88 citations


Journal ArticleDOI
TL;DR: This work tries to show that structuring classifiers into relevant multistage organisations can widen this boundary, as well as the limits of majority voting error, even more, and investigates the sensitivity of boundary distributions of classifier outputs to small discrepancies modelled by the random changes of votes.
Abstract: A robust character of combining diverse classifiers using a majority voting has recently been illustrated in the pattern recognition literature. Furthermore, negatively correlated classifiers turned out to offer further improvement of the majority voting performance even comparing to the idealised model with independent classifiers. However, negatively correlated classifiers represent a very unlikely situation in real-world classification problems, and their benefits usually remain out of reach. Nevertheless, it is theoretically possible to obtain a 0% majority voting error using a finite number of classifiers at error levels lower than 50%. We attempt to show that structuring classifiers into relevant multistage organisations can widen this boundary, as well as the limits of majority voting error, even more. Introducing discrete error distributions for analysis, we show how majority voting errors and their limits depend upon the parameters of a multiple classifier system with hardened binary outputs (correct/incorrect). Moreover, we investigate the sensitivity of boundary distributions of classifier outputs to small discrepancies modelled by the random changes of votes, and propose new more stable patterns of boundary distributions. Finally, we show how organising classifiers into different structures can be used to widen the limits of majority voting errors, and how this phenomenon can be effectively exploited.

67 citations


Journal ArticleDOI
TL;DR: This paper explores the advantages of combining the MASKS and MKL-based classifiers, which are specifically designed for the fingerprint classification task, and proposes a combination at the ‘abstract level’ and a fusion at the 'measurement level' for continuous classification.
Abstract: Fingerprint classification is a challenging pattern recognition problem which plays a fundamental role in most of the large fingerprint-based identification systems. Due to the intrinsic class ambiguity and the difficulty of processing very low quality images (which constitute a significant proportion), automatic fingerprint classification performance is currently below operating requirements, and most of the classification work is still carried out manually or semi-automatically. This paper explores the advantages of combining the MASKS and MKL-based classifiers, which we have specifically designed for the fingerprint classification task. In particular, a combination at the ‘abstract level’ is proposed for exclusive classification, whereas a fusion at the ‘measurement level’ is introduced for continuous classification. The advantages of coupling these distinct techniques are well evident; in particular, in the case of exclusive classification, the FBI challenge, requiring a classification error ≤ 1% at 20% rejection, was met on NIST-DB14.

56 citations


Journal ArticleDOI
TL;DR: The weighted mean of a pair of strings is introduced, formal properties of the weighted mean are shown, a procedure for its computation is described, and practical examples are given.
Abstract: String matching and string edit distance are fundamental concepts in structural pattern recognition. In this paper, the weighted mean of a pair of strings is introduced. Given two strings, x and y, where d(x, y) is the edit distance of x and y, the weighted mean of x and y is a string z that has edit distances d(x, z) and d(z, y)to x and y, respectively, such that d(x, z) _ d(z, y) = d(x, y). We'll show formal properties of the weighted mean, describe a procedure for its computation, and give practical examples.

44 citations


Journal ArticleDOI
TL;DR: An optimisation scheme which includes several steps and assures a convergence to a useful solution is introduced for constructing and training a hybrid architecture of projection-based units and radial basis functions.
Abstract: We introduce a mechanism for constructing and training a hybrid architecture of projection-based units and radial basis functions. In particular, we introduce an optimisation scheme which includes several steps and assures a convergence to a useful solution. During network architecture construction and training, it is determined whether a unit should be removed or replaced. The resulting architecture often has a smaller number of units compared with competing architectures. A specific overfitting resulting from shrinkage of the RBF radii is addressed by introducing a penalty on small radii. Classification and regression results are demonstrated on various benchmark data sets and compared with several variants of RBF networks [1,2]. A striking performance improvement is achieved on the vowel data set [3].

44 citations


Journal ArticleDOI
TL;DR: A neural architecture, based on several self-organising maps, is presented which counteracts the parameter drift problem for an array of conducting polymer gas sensors when used for odour sensing and shows long-term stable behaviour.
Abstract: A neural architecture, based on several self-organising maps, is presented which counteracts the parameter drift problem for an array of conducting polymer gas sensors when used for odour sensing. The neural architecture is named mSom, where m is the number of odours to be recognised, and is mainly constituted of m maps; each one approximates the statistical distribution of a given odour. Competition occurs both within each map and between maps for the selection of the minimum map distance in the Euclidean space. The network (mSom) is able to adapt itself to new changes of the input probability distribution by repetitive self-training processes based on its experience. This architecture has been tested and compared with other neural architectures, such as RBF and Fuzzy ARTMAP. The network shows long-term stable behaviour, and is completely autonomous during the testing phase, where re-adaptation of the neurons is needed due to the changes of the input probability distribution of the given data set.

Journal ArticleDOI
TL;DR: The trim and spread combiners are analysed, both based on linear combinations of the ordered classifier outputs, and it is shown that in the presence of uneven classifier performance, they often provide substantial gains over both linear and simple order statistics combiners.
Abstract: Integrating the outputs of multiple classifiers via combiners or meta-learners has led to substantial improvements in several difficult pattern recognition problems. In this article, we investigate a family of combiners based on order statistics, for robust handling of situations where there are large discrepancies in performance of individual classifiers. Based on a mathematical modelling of how the decision boundaries are affected by order statistic combiners, we derive expressions for the reductions in error expected when simple output combination methods based on the median, the maximum and in general, the ith order statistic, are used. Furthermore, we analyse the trim and spread combiners, both based on linear combinations of the ordered classifier outputs, and show that in the presence of uneven classifier performance, they often provide substantial gains over both linear and simple order statistics combiners. Experimental results on both real world data and standard public domain data sets corroborate these findings.

Journal ArticleDOI
TL;DR: A feature-based approach to the serial multi-stage combination of classifiers is shown empirically to provide a substantial decrease in the computational complexity, while maintaining the accuracy level of a single-stage classifier or even improving it.
Abstract: A new approach to the serial multi-stage combination of classifiers is proposed. Each classifier in the sequence uses a smaller subset of features than the subsequent classifier. The classification provided by a classifier is rejected only if its decision is below a predefined confidence level. The approach is tested on a two-stage combination of k-Nearest Neighbour classifiers. The features to be used by the first classifier in the combination are selected by two stand-alone algorithms (Relief and Info-Fuzzy Network, or IFN) and a hybrid method, called `IFN + Relief'. The feature-based approach is shown empirically to provide a substantial decrease in the computational complexity, while maintaining the accuracy level of a single-stage classifier or even improving it.

Journal ArticleDOI
TL;DR: This paper focuses on presenting state-of-the-art systems based on the fusion of boundary/surface-based with region-based techniques, also called regional-geometric deformation models, which takes the paradigm of partial differential equations in the level set framework.
Abstract: Extensive growth in functional brain imaging, perfusion-weighted imaging, diffusion-weighted imaging, brain mapping and brain scanning techniques has led tremendously to the importance of cerebral cortical segmentation both in 2-D and 3-D from volumetric brain magnetic resonance imaging data sets. Besides that, recent growth in deformable brain segmentation techniques in 2-D and 3-D has brought the engineering community, such as the areas of computer vision, image processing, pattern recognition and graphics, closer to the medical community, such as to neuro-surgeons, psychiatrists, oncologists, neuro-radiologists and internists. In Part I of this research (see Suri et al [1]), an attempt was made to review the state-of-the-art in 2-D and 3-D cerebral cortical segmentation techniques from brain magnetic resonance imaging based on two main classes: region- and boundary/surface-based. More than 18 different techniques for segmenting the cerebral cortex from brain slices acquired in orthogonal directions were shown using region-based techniques. We also showed more than ten different techniques to segment the cerebral cortex from magnetic resonance brain volumes using boundary/surface-based techniques. This paper (Part II) focuses on presenting state-of-the-art systems based on the fusion of boundary/surface-based with region-based techniques, also called regional-geometric deformation models, which takes the paradigm of partial differential equations in the level set framework. We also discuss the pros and cons of these various techniques, besides giving the mathematical foundations for each sub-class in the cortical taxonomy. Special emphasis is placed on discussing the advantages, validation, challenges and neuro-science/clinical applications of cortical segmentation.

Journal ArticleDOI
TL;DR: The proposed moderation method improves the performance of the multiple classifier system significantly and can be minimised by marginalising the k-NN estimates using the Bayesian prior.
Abstract: The performance of a multiple classifier system combining the soft outputs of k-Nearest Neighbour (k-NN) Classifiers by the product rule can be degraded by the veto effect. This phenomenon is caused by k-NN classifiers estimating the class a posteriori probabilities using the maximum likelihood method. We show that the problem can be minimised by marginalising the k-NN estimates using the Bayesian prior. A formula for the resulting moderated k-NN estimate is derived. The merits of moderation are examined on real data sets. Tests with different bagging procedures indicate that the proposed moderation method improves the performance of the multiple classifier system significantly.

Journal ArticleDOI
TL;DR: It is shown that the proposed parallel fusion of probabilistic neural networks produces biologically plausible structures and improves the resulting recognition performance.
Abstract: The main motivation of this paper is to design a statistically well justified and biologically compatible neural network model and, in particular, to suggest a theoretical interpretation of the well known high parallelism of biological neural networks. We consider a novel probabilistic approach to neural networks developed in the framework of statistical pattern recognition, and refer to a series of theoretical results published earlier. It is shown that the proposed parallel fusion of probabilistic neural networks produces biologically plausible structures and improves the resulting recognition performance. The complete design methodology based on the EM algorithm has been applied to recognise unconstrained handwritten numerals from the database of Concordia University Montreal. We achieved a recognition accuracy of about 95%, which is comparable with other published results.

Journal ArticleDOI
TL;DR: A new three-stage verification system which is based on three types of features: global features; local features of the corner points; and function features that contain information of each point of the signatures is presented.
Abstract: This paper presents a new three-stage verification system which is based on three types of features: global features; local features of the corner points; and function features that contain information of each point of the signatures. The first verification stage implements a parameter-based method, in which the Mahalanobis distance is used as a dissimilarity measure between the signatures. The second verification stage involves corner extraction and corner matching. It also performs signature segmentation. The third verification stage implements a function-based method, which is based on an elastic matching algorithm establishing a point-to-point correspondence between the compared signatures. By combining the three different types of verification, a high security level can be reached. According to our experiments, the rates of false rejection and false acceptance are, respectively, 5.8% and 0%.

Journal ArticleDOI
TL;DR: An ideal weighted sampling method, where the pixels in the image lying at the intersection of sensor cells, are subdivided into smaller sub-pixels, and an interpolation method using a variable width interpolation mask, whose size varies exponentially with the size and shape of the cells in the sensor array are introduced.
Abstract: Space-variant imaging sensors have many advantages over conventional raster imaging sensors. They provide a large field of view for a given pixel count while maintaining a high resolution at the centre of the field of view and, in addition, produce a mapping that is scale and rotation invariant. The effectiveness of the sensor depends greatly upon the geometry used and the sampling methods employed. In this paper, we define a sensor geometry and introduce an ideal weighted sampling method, where the pixels in the image lying at the intersection of sensor cells, are subdivided into smaller sub-pixels, and an interpolation method using a variable width interpolation mask, whose size varies exponentially with the size and shape of the cells in the sensor array. We compare the computational requirements of these methods, and show that they are scale and rotation invariant, when the image is scaled or rotated about its centre, giving the sensor a functionality similar to that provided by the retinal mapping present in the mammalian retina. These results illustrate the advantages that can be obtained in real-time tracking applications in computer vision, where computational and memory requirements need to be kept to a minimum.

Journal ArticleDOI
TL;DR: A motion segmentation method useful for representing efficiently a video shot as a static mosaic of the background plus sequences of the objects moving in the foreground generates an MPEG-4 compliant, layered representation useful for video coding, editing and indexing.
Abstract: This paper presents a motion segmentation method useful for representing efficiently a video shot as a static mosaic of the background plus sequences of the objects moving in the foreground. This generates an MPEG-4 compliant, layered representation useful for video coding, editing and indexing. First, a mosaic of the static background is computed by estimating the dominant motion of the scene. This is achieved by tracking features over the video sequence and using a robust technique that discards features attached to the moving objects. The moving objects get removed in the final mosaic by computing the median of the grey levels. Then, segmentation is obtained by taking the pixelwise difference between each frame of the original sequence and the mosaic of the background. To discriminate between the moving object and noise, temporal coherence is exploited by tracking the object in the binarised difference image sequence. The automatic computation of the mosaic and the segmentation procedure are illustrated with real sequences experiments. Examples of coding and content-based manipulation are also shown.

Journal ArticleDOI
TL;DR: From numerical experiments involving both regression and classification problems, the OLA was shown to provide better generalisation performance than simple committee, boosting and bagging approaches, when insufficient and noisy training data are given.
Abstract: We propose Observational Learning Algorithm (OLA), an ensemble learning algorithm with T and O steps alternating. In the T-step, an ensemble of networks is trained with a training data set. In the O-step, ‘virtual’ data are generated in which each target pattern is determined by observing the member networks’ output for the input pattern. These virtual data are added to the training data and the two steps are repeatedly executed. The virtual data was found to play the role of a regularisation term as well as that of temporary hints having the auxiliary information regarding the target function extracted from the ensemble. From numerical experiments involving both regression and classification problems, the OLA was shown to provide better generalisation performance than simple committee, boosting and bagging approaches, when insufficient and noisy training data are given. We examined the characteristics of the OLA in terms of ensemble diversity and robustness to noise variance. The OLA was found to balance between ensemble diversity and the average error of individual networks, and to be robust to the variance of noise distribution. Also, OLA was applied to five real world problems from the UCI repository, and its performance was compared with bagging and boosting methods.

Journal ArticleDOI
TL;DR: A new approach to colour image morphology is proposed, based on a new ordering of vectors in the HSV colour space that is partial ordering, which is compatible to the standard greyscale morphology.
Abstract: The extension of concepts of greyscale morphology to colour image processing requires the use of a proper ordering of vectors (colours) and the definitions of infimum and supremum operators in an appropriate colour space. In this paper, a new approach to colour image morphology is proposed. It is based on a new ordering of vectors in the HSV colour space that is partial ordering. The proposed approach is hue preserving, and it is not a component-wise technique. Its basic characteristic is that it is compatible to the standard greyscale morphology: its fundamental and secondary operations possess the same basic properties as their greyscale counterparts, and furthermore, it is identical to greyscale morphology when it is applied to greyscale images. Examples that illustrate the application of the defined operations to colour images are provided. Moreover, the usefulness of the new method in various colour image processing applications, such as colour image edge detection, object recognition, vector top-hat filtering and skeleton extraction, is demonstrated.

Journal ArticleDOI
TL;DR: This article deals with the combination of pattern classifiers with two reject options, and proposes to combine the first steps of these classifiers using concepts from the theory of evidence to reject classes before using the combination rule.
Abstract: This article deals with the combination of pattern classifiers with two reject options. Such classifiers operate in two steps and differ on the managing of ambiguity and distance rejection (independently or not). We propose to combine the first steps of these classifiers using concepts from the theory of evidence. We propose some intelligent basic probability assignment to reject classes before using the combination rule. After combination, a decision rule is proposed for classifying or rejecting patterns either for distance or for ambiguity. We emphasize that rejection is not related to a lack of consensus between the classifiers, but to the initial reject options. In the case of ambiguity rejection, a class-selective approach has been used. Some illustrative results on artificial and real data are given.

Journal ArticleDOI
TL;DR: 1. IBM.
Abstract: 1. IBM. Data mining: Make your data work for you like never before, http://direct.boulder.ibm.com/bi/tech/mining 2. Geller J, Kitano H, Suttner CB (editors). Parallel Processing for Artificial Intelligence S. Elsevier, 1997 3. Kitano H, Hendler J (editors). Massively Parallel Artificial Intelligence. AAAI Press, 1994 4. Bond A, Gasser L (editors). Readings in Distributed Artificial Intelligence. Morgan Kaufmann, 1988 5. Weiss G (editor). Multiagent Systems: A Modern Approach to Distributed Artificial Intelligence. MIT Press, 1999 6. Han J, Kamber M. Data Mining: Concepts and Techniques. Morgan Kaufmann, 2001 7. Dowd K, Severance C. High Performance Computing. O’Reilly, 1998 DIANNE J. COOK

Journal ArticleDOI
TL;DR: A survey of Multiple Classifier Systems from an MCS perspective opens up new possibilities within MCS as well as provides new formal bases for the central underlying ideas, such as classifier independence and diversity.
Abstract: One vigorous branch of research aimed at improving the performance of pattern recognition systems explores the possibilities for exploiting the differences between a set of variously configured classifiers. This is the field of Multiple Classifier Systems (MCS), and it is based on the premise that it ought to be possible to organise and exploit the strengths and weaknesses of individual classifiers such that the MCS performance is superior to that of any of its components. Important concerns are the efficiency of multiple classifier construction, and the effectiveness of the final MCS. What property or properties of the set of multiple classifiers are being exploited by the various decision strategies, and how are the desired properties to be realised within a set of classifiers? Analogous ideas and strands of research have arisen within both software engineering and neural computing. This paper surveys these other two fields from an MCS perspective with the goal of revealing useful results that should have direct application for current work in MCS. In particular, the survey opens up new possibilities within MCS as well as provides new formal bases for the central underlying ideas, such as classifier independence and diversity. The exploration of diversity is extended to a consideration of MCSs in which the component classifiers are specialised for classification of an identifiable subset of the complete classification problem. Results are given of an empirical study of an automatic specialisation strategy that demonstrates the predictive use of several diversity measures. Finally, a taxonomy is presented as a unifying framework for the many varieties of MCSs.

Journal ArticleDOI
TL;DR: A combination scheme labelled ‘Bagfs’, in which new learning sets are generated on the basis of both bootstrap replicates and random subspaces, shows that on average, Bagfs exhibits the best agreement between prediction and supervision.
Abstract: Several ways of manipulating a training set have shown that weakened classifier combination can improve prediction accuracy. In the present paper, we focus on learning set sampling (Breiman's Bagging) and random feature subset selections (Ho's Random Subspaces). We present a combination scheme labelled 'Bagfs', in which new learning sets are generated on the basis of both bootstrap replicates and random subspaces. The performances of the three methods (Bagging, Random Subspaces and Bagfs) are compared to the standard Adaboost algorithm. All four methods are assessed by means of a decision-tree inducer (C4.5). In addition, we also study whether the number and the way in which they are created has a significant influence on the performance of their combination. To answer these two questions, we undertook the application of the McNemar test of significance and the Kappa degree-of-agreement. The results, obtained on 23 conventional databases, show that on average, Bagfs exhibits the best agreement between prediction and supervision.

Journal ArticleDOI
TL;DR: A series of standard multi-dimensional feature analysis techniques are shown to improve the classification accuracy of the dynamic properties of task execution, and hence the sensitivity to the detection of neglect and the validity of this novel application.
Abstract: Visuo-spatial neglect is recognised as a major barrier to recovery following a stroke or head injury. A standard clinical assessment technique to assess the condition is a pencil-and-paper based cancellation task. Traditional static analysis of this task involves counting the number of targets correctly cancelled on the test sheet. Using a computer-based test capture system, this paper presents the novel application of using a series of standard pattern recognition techniques to examine the diagnostic capability of a number of dynamic features relating to the sequence in which the targets were cancelled. While none of the individual dynamic features is as sensitive to neglect as the conventional static analysis, a series of standard multi-dimensional feature analysis techniques are shown to improve the classification accuracy of the dynamic properties of task execution, and hence the sensitivity to the detection of neglect and the validity of this novel application. Combining the outcome of the dynamic sequence-based features with the conventional static analysis further improves the overall sensitivity of the two cancellation tasks included in this study. The algorithmic nature of the methodology for feature extraction objectively and consistently assesses patients, thereby improving the repeatability of the task.

Journal ArticleDOI
TL;DR: Experimental results are given to show that the new affine invariants are less sensitive to noise and the recognition rate is increased when using both the available spatial domain and the proposed frequency domain affine moment invariants.
Abstract: In this paper, a new set of affine moment invariants is proposed in the frequency domain. By thresholding the magnitude of the Discrete Fourier Transform (DFT) of affine transformation-related images, new images which are also related by affine transformation are constructed. Then some affine invariant features in the frequency domain are obtained from the reconstructed images. Experimental results are given to show that the new affine invariants are less sensitive to noise and the recognition rate is increased when using both the available spatial domain and the proposed frequency domain affine moment invariants.

Journal ArticleDOI
TL;DR: A broad class of models named Generalised Additive Multi-Mixture Models (GAM-MM), based on a multiple combination of mixtures of classifiers to be used in both the regression and classification cases, is presented.
Abstract: In the context of supervised statistical learning, we present a broad class of models named Generalised Additive Multi-Mixture Models (GAM-MM), based on a multiple combination of mixtures of classifiers to be used in both the regression and classification cases. In particular, we additively combine mixtures of different types of classifiers, defining an ensemble composed of nonparametric tools (tree-based methods), semiparametric tools (scatterplot smoothers) and parametric tools (linear regression). Within this approach, we define a classifier scoring criterion to be jointly used with the bagging procedure for estimation of the mixing parameters, and describe the GAM-MM estimation procedure, that adaptively works by iterating a backfitting-like algorithm and a local scoring procedure until convergence. The effectiveness of our approach in modelling complex data structures is evaluated by presenting the results of some applications on real and simulated data.

Journal ArticleDOI
Rae-Hong Park1
TL;DR: This paper presents complex-valued feature masks formulated by directional filtering of 3 × 3 feature masks (e.g. Prewitt, Sobel, Frei-Chen, Kirsch and roof masks), with the different number of directions N (= 8, 4 and 2).
Abstract: A two-dimensional (2D) N-directional edge detection filter was represented as a pair of real masks, i.e. by one complex-valued matrix [1]. The 3 × 3 compass gradient edge masks have often been used because of their simplicity, where N compass masks are constructed by rotating the kernel mask by an incremental angle of 2ź/N. This paper presents complex-valued feature masks formulated by directional filtering of 3 × 3 feature masks (e.g. Prewitt, Sobel, Frei-Chen, Kirsch and roof masks), with the different number of directions N (= 8, 4 and 2). The same concept can be applied to any types of filtering/masks with arbitrary N.

Journal ArticleDOI
TL;DR: An algorithm for 3D-object reconstruction from the vectorised and parameterised drawing is proposed, based on the detection of volumetric solid-state object components (primitives), and performing theoretic-set operations with the components.
Abstract: In this paper, the problem of 3D-object model reconstruction from engineering drawing projections is analysed, and its main stages are shown. Image vectorisation and entity recognition is mentioned briefly, the main focus being editing or the parameterisation of vectorised drawings and 3D object model reconstruction from vectorised ED projections. Vectorised drawing, as a rule, do not exactly correspond to sizes and other features (touching, parallelity, perpendicularly, symmetry, collinearity, etc.) being available on the initial drawing, and this ED vector model is not suitable for direct use in CAD systems. That is why the parameterisation stage is introduced and considered in detail. An algorithm for 3D-object reconstruction from the vectorised and parameterised drawing is proposed. The algorithm is based on the detection of volumetric solid-state object components (primitives), and performing theoretic-set operations with the components. Practical experience in realising these stages is shown.