In the framework of Alzheimer’s disease prediction systems, it is widely agreed that handwriting seems to be one of the first skills to be influenced by the onset of such a disease. In the large majority of cases, the above systems consider information relating to the dynamics of the handwriting process, directly derived from online handwriting samples. This kind of features, however, are not able to capture the alterations in the shape, size and thickness of the handwritten traits, which may be produced by the alterations in motor control due to neurodegenerative disorders. Following this line of thought, in a previous study we combined shape and dynamic information by generating synthetic color images from online handwriting samples, where the color of each elementary trait encodes, in the three RGB channels, the dynamic information associated to that trait. Finally, we exploited the capability of Deep Neural Networks to automatically extract features from raw images, following the Transfer Learning approach. The results obtained with this approach did not show significant improvements compared to those obtained with the use of dynamic information only, probably because approximating the original traits with straight lines of predefined thickness results in a loss of information on their actual shape and thickness. Moving from these considerations, the purpose of our study is to verify whether automatically extracting features directly from offline handwriting images, thus considering the original shape of the handwritten trace, could provide better results. Again, we exploited the capability of Deep Neural Networks to automatically extract features from raw images. The preliminary experimental results confirmed the effectiveness of the proposed approach.

Early diagnosis of neurodegenerative disorders, such as Alzheimer's Disease (AD), is very important to reduce their effects and to improve both quality and life expectancy of patients. In this context, it is generally agreed that handwriting is one of the first skills altered by the onset of AD. For this reason, the analysis of handwriting and the study of its alterations has become of great interest in order to formulate the diagnosis as soon as possible. A fundamental aspect for the use of these techniques is the definition of effective features, which allows the system to distinguish the natural alterations of handwriting due to age, from those caused by neurodegenerative disorders. Starting from these considerations, the aim of our study is to verify whether the combined use of both shape and dynamic features allows a decision support system to improve performance for AD diagnosis. To this purpose, starting from a database of on-line handwriting samples, we generated for each of them an off-line synthetic color image, where the color of each elementary trait encodes, in the three RGB channels, the dynamic information associated with that trait. To verify the role played by dynamic information, we also generated simple binary images, containing only shape information. Finally, we exploited the ability of Convolutional Neural Network (CNN) to automatically extract features on both color and binary images. The experimental results have confirmed that dynamic information allows a performance improvement with respect to the binary images.

Early detection of Alzheimer's Disease (AD) is essential in order to initiate therapies that can reduce the effects of such a disease, improving both life quality and life expectancy of patients. Among all the activities carried out in our daily life, handwriting seems one of the first to be influenced by the arise of neurodegenerative diseases. For this reason, the analysis of handwriting and the study of its alterations has become of great interest in this research field in order to make a diagnosis as early as possible. In recent years, many studies have tried to use classification algorithms applied to handwriting to implement decision support systems for AD diagnosis. A key issue for the use of these techniques is the detection of effective features, that allow the system to distinguish the natural handwriting alterations due to age, from those caused by neurodegenerative disorders. In this context, many interesting results have been published in the literature in which the features have been typically selected by hand, generally considering the dynamics of the handwriting process in order to detect motor disorders closely related to AD. Features directly derived from handwriting generation models can be also very helpful for AD diagnosis. It should be remarked, however, that the above features do not consider changes in the shape of handwritten traces, which may occur as a consequence of neurodegenerative diseases, as well as the correlation among shape alterations and changes in the dynamics of the handwriting process. Moving from these considerations, the aim of this study is to verify if the combined use of both shape and dynamic features allows a decision support system to improve performance for AD diagnosis. To this purpose, starting from a database of on-line handwriting samples, we generated for each of them a synthetic off-line colour image, where the colour of each elementary trait encodes, in the three RGB channels, the dynamic information associated to that trait. Finally, we exploited the capability of Deep Neural Networks (DNN) to automatically extract features from raw images, following the Transfer Learning approach. The experimental comparison of the results obtained by using standard features and features extracted according the above procedure, confirmed the effectiveness of our approach.

Structure learning is a very important problem in the context of Bayesian networks (BNs). For this reason, it has been largely studied in the last few years and many approaches have been presented to find an optimal structure based on training samples. In a previous paper, we proposed an evolutionary algorithm for BN structure learning based on a data structure specifically devised for encoding BNs. The proposed approach was used to combine the responses of classifier ensembles. In this paper, we present a further improvement along this direction, in that we have developed a novel mutation operator that allows a more effective exploration of the search space. The devised operator is able to modify both node ordering and the connection topology of the encoded BN. The experimental results, obtained by using five benchmark datasets, confirmed the effectiveness of the proposed approach.

We propose a model for estimating the time to transcribe a large collection of historical handwritten documents when the transcription is assisted by a keyword spotting system following the query-by-string approach. The model assumes that the system is segmentation-based and provides as output the transcription of each item (either right or wrong) or a reject. We also assume that any other information the system may need is obtained from the training set. The model has been validated by comparing its estimates with the actual time required for the manual transcription of pages from the Bentham dataset. Eventually, we discuss possible ways of extending the model to consider different kind of keyword spotting system, such as those providing the output in terms of a ranked list of alternatives and/or adopting the query-by-example approach.

We propose a quantitative approach to both feature evaluation and comparison that combines Forensic Handwriting Examination best practices with Pattern Recognition methodologies. The former provide a set of features that are meant to capture the distinctive aspects of handwriting, the latter the computational tools for the quantitative evaluation of the features values as well as for their comparison. We will show that such a combined approach leads to a procedure that is theoretically sounds and can be expressed in terms the document examiners are familiar with. Eventually, we will suggest possible ways of using the results of the proposed approach in forensic handwriting examiners casework.

The form processing systems commercially available include a verification step during which a human operator verifies the output provided by the system to ensure 100% accuracy. In order to reduce the time and the cost of such a stage, the OCR engine incorporated into the system provides a reliability measure of the classification to be used for implementing a reject option: in this way only rejected samples are passed to the verification stage. Most of the strategies for designing such a reject option consider that the source of classification errors are within the OCR engine. Such an assumption becomes less reasonable as the forms become less structured, as in case when boxes are provided for the entire data field and not only for isolated characters. Under these circumstances, we investigate to which extent the reliability measure provided by an OCR engine designed to deal with boxed isolated characters can be used to detect both segmentation and classification errors. The experimental results, obtained on a large data set of forms currently in use by a large organization, show that the proposed method successfully achieves its aim. It represents a powerful tool for the system manager to plan system enhancement as the volume of forms containing less constrained data fields increases.

In many real world applications, the cost of a wrong decision may be much higher than the benefit of having a high classification rate. This kind of classification problems represent very challenging tasks, because they require highly reliable systems and may benefit of introducing a reject option. To this purpose, many classification systems have been proposed, among which classifier ensembles represent a successful example. This approach aims at combining classifiers making uncorrelated errors. In this framework, the Random Forest (RF) represents a case of special interest. A RF is made of a suitable ensemble of decision trees and has proved to be very effective in several fields. In this paper, the RF classification reliability is experimentally analyzed with reference to the case of handwriting recognition. The aim is to verify if such reliability can be effectively used to introduce a new reject option, which considers, for each unknown sample, a subset of few classes including with high probability the correct one. The whole system operates as a pre-classification stage and the reject option should allow us to obtain a low error rate without significantly affecting the recognition rate. Experiments, carried out on two real world datasets, have shown the effectiveness of the proposed method.

Classifier combination methods have proved to be an effective tool for increasing the performance in pattern recognition applications. The rationale of this approach follows from the observation that appropriately diverse classifiers make uncorrelated errors. Unfortunately, this theoretical assumption is not easy to satisfy in practical cases, thus reducing the performance obtainable with any combination strategy. In this paper we propose a new weighted majority vote rule which try to solve this problem by jointly analyzing the responses provided by all the experts, in order to capture their collective behavior when classifying a sample. Our rule associates a weight to each class rather than to each expert and computes such weights by estimating the joint probability distribution of each class with the set of responses provided by all the experts in the combining pool. The probability distribution has been computed by using the \textit{naive Bayes} probabilistic model. Despite its simplicity, this model has been successfully used in many practical applications, often competing with much more sophisticated techniques. The experimental results, performed by using three standard databases of handwritten digits, confirmed the effectiveness of the proposed method.

We present a method for off-line reading of cursive handwriting, which derives from modelling handwriting as a complex movement. The method includes a step for recovering the writing order from static images of handwriting, a segmentation algorithm that decomposes the “unfolded” ink into strokes, an ink matching step to compare the ink of the unknown handwriting with those of a set of reference words, of whom the transcripts are given, and a graph search algorithm to search for the best interpretation among the possible ones. The method does not involve any feature extraction, nor a classification stage and may benefit from a linguistic context, if available. We report the results of experiments on 8,000 samples, draw some conclusions and outline further developments.

We propose a feature selection--based approach for improving classification performance of a two stage classification system in contexts where a high number of features is involved. A problem with a set of $N$ classes is subdivided into a set of $N$ two class problems. In each problem, a GA--based feature selection algorithm is used for finding the best subset of features. These subsets are then used for training $N$ classifiers. In the classification phase, unknown samples are given in input to each of the trained classifiers by using the corresponding subspace. In case of conflicting responses, the sample is sent to a suitably trained supplementary classifier. The proposed approach has been tested on a real world dataset containing hyper--spectral image data. The results favourably compare with those obtained by other methods on the same data.

We present a method to recover the dynamic writing order from static images of handwriting. The static handwriting is initially represented by its skeleton, which is then converted into a graph, whose arcs correspond to the skeleton branches, and nodes to either end point or branch point of the skeleton. Criteria derived by handwriting generation are then applied to transform the graph in such a way that all its nodes, but the first and the last, have an even degree, so that it can be traversed from the first to the last by using the Fleury's algorithm. The experimental results show that combining criteria derived from handwriting generation models with graph traversal leads to reconstruct the original sequence produced by a writer even in case of complex handwriting, i.e handwriting with retracing, crossings and pen-up's.

Combining classifier methods have shown their effectiveness in a number of applications. Nonetheless, using simultaneously multiple classifiers may result in some cases in a reduction of the overall performance, since the responses provided by some of the experts may generate consensus on a wrong decision even if other experts provided the correct one. To reduce these undesired effects, in a previous study, we proposed a combining method based on the use of a Bayesian Network. In this paper, we present an improvement of that method which allows to solve some of the drawbacks exhibited by standard learning algorithms for Bayesian Networks. The proposed method is based on an Evolutionary Algorithm which uses a specifically devised data structure to encode direct acyclic graphs. This data structure allows to effectively implement crossover and mutation operators. The experimental results, obtained by using three standard databases, confirmed the effectiveness of the method.

In the field of handwriting recognition, classifier combination received much more interest than the study of powerful individual classifiers. This is mainly due to the enormous variability among the patterns to be classified, that typically requires the definition of complex high dimensional feature spaces: as the overall complexity increases, the risk of inconsistency in the decision of the classifier increases as well. In this framework, we propose a new combining method based on the use of a Bayesian Network. In particular, we suggest to reformulate the classifier combination problem as a pattern recognition one in which each input pattern is associated to a feature vector composed by the output of the classifiers to be combined. A Bayesian Network is then used to automatically infer the probability distribution for each class and eventually to perform the final classification. Experiments have been performed by using two different pools of classifiers, namely an ensemble of Learning Vector Quantization neural networks and an ensemble of Back Propagation neural networks, and handwritten specimen from the UCI Machine Learning Repository. The obtained performance has been compared with those exhibited by multi-classifier systems adopting the classifiers, but three of the most effective and widely used combining rules: the Majority Vote, the Weighted Majority Vote and the Borda Count.

A large amount of handwritten documents exist in image form, as scanned documents. The supporting electronic media allows for better preservation, but to access their content they must be processed by some kind of recognition technologies that convert the image to searchable text. In case of cursively written documents, even the best available technology introduces recognition errors that may drive down the performance of a document retrieval system. We propose a recognition-free approach which embodies two main components: a shape matching algorithm, working on the ink, and a string matching algorithm working on the ink interpretation of a reference set. Experiments on a data set of 16,500 cursive words produced by hundreds of writers show promising results and suggest that the proposed method can be a viable tool to build inexpensive retrieval system for cursive documents.

We present a Genetic Algorithm based feature selection approach according to which feature subsets are represented by individuals of an evolving population. Evolution is controlled by a fitness function taking into account statistical properties of the input data in the subspace represented by each individual, and aims to select the smallest feature subset that optimizes class separability. The originality of our method lies particularly in the definition of the evaluation function. The proposed approach has been tested on a standard database of handwritten digits, showing to be effective both for reducing the number of features used and for improving classifier performance.

We propose a new feature selection algorithm for remote sensing image classification. Our approach has been especially devised for applications in which there is a large number of different features that can be potentially selected, implying that the search space is complex and high-dimensional. In this framework, our proposal is that of reformulating the feature selection problem as the search for the optimal subspace in which the different classes are more effectively discriminated. The search has been performed by using a genetic algorithm in which each individual encode the choice of a subspace, and its fitness is a measure of the class seperability in that subspace. The experimental results, performed on two databases, confirmed the effectiveness of the approach.

A new prototyping method based on the evolutionary computation paradigm and on the concept of Vector Quantization is proposed. It uses a specifically devised evolutionary algorithm for evolving a set of prototype feature vectors and does not require any a priori knowledge about either the actual number of prototypes or the statistical properties of the input data. Experiments performed by using both synthetic data and handwritten digits randomly extracted from the NIST database have confirmed the effectiveness of the approach.

In the framework of a saliency-based approach for segmenting cursive handwriting into elementary strokes, we propose a smoothing technique based on the use of a Wavelet Transform to describe the electronic ink at different resolutions. According to such an approach, derived in analogy with those proposed in the literature for early visual tasks in primates, curvature maxima corresponding to actual segmentation points are separated from those produced by the different source of noise affecting the handwriting generation process. This is obtained by evaluating the curvature maxima at different levels of resolution, and by arranging them into a pyramidal structure, from which a saliency map is eventually achieved by combining the values across all possible scales. Such maps enjoy the property of exhibiting higher values in correspondence of regions of the original ink where curvature maxima survive along difference scales, thus indicating that those regions should correspond to the parts of the ink where two strokes join. Experiments performed by using both standard databases and one of online words collected in our laboratory, have shown high performance in terms of correct segmentation, stability and predictability of the results.

The problem of locating all the optima within a multimodal fitness landscape has been widely addressed in evolutionary computation, and many solutions, based on a large variety of different techniques, have been proposed in the literature. Among them, fitness sharing (FS) is probably the best known and the most widely used. The main criticisms to FS concern both the lack of an explicit mechanism for identifying or providing any information about the location of the peaks in the fitness landscape, and the definition of species implicitly assumed by FS. We present a mechanism of FS, i.e., dynamic fitness sharing, which has been devised in order to overcome these limitations. The proposed method allows an explicit, dynamic identification of the species discovered at each generation, their localization on the fitness landscape, the application of the sharing mechanism to each species separately, and a species elitist strategy. The proposed method has been tested on a set of standard functions largely adopted in the literature to assess the performance of evolutionary algorithms on multimodal functions. Experimental results confirm that our method performs significantly better than FS and other methods proposed in the literature without requiring any further assumption on the fitness landscape than those assumed by the FS itself.

We introduce a multiple classifier system that incorporates a global optimization technique based on a Breeder Genetic Algorithm for dynamically selecting the set of experts to be included in the pool. The proposed technique is applicable when the experts provide both the class assigned to the input sample and a measure of the reliability of the classification. For each sample, the experts selected for participating in the voting rule are those whose reliability is larger than a given threshold. There are as many thresholds as the number of experts by the number of classes. The values of the thresholds aimed at selecting the best set of experts for each input sample are determined by the Breeder Genetic Algorithm. The reliability measures provided by the experts of the pool are also used to implement the tie-break mechanism needed within the voting scheme. The system has been tested on the Image database from the UCI database repository by using as classifiers an ensemble of Back-Propagation neural network and an ensemble of Learning Vector Quantization neural network. The voting schemes adopted are the Majority Vote, the Weighted Majority Vote and the Borda Count. The performance of the system is compared with those exhibited by the multi-expert systems exploiting the same combining rules without the dynamic selection of the expert.

We propose a genetic programming based approach for generating prototypes in a classification problem. In this context, the set of prototypes to which the samples of a data set can be traced back is coded by a multitree, i.e. a set of trees, which represents the chromosome. Differently from other approaches, our chromosomes are of variable length. This allows coping with those classification problems in which one or more classes consist of subclasses. The devised approach has been tested on several problems and the results compared with those obtained by a different genetic programming based approach recently proposed in the literature.

We propose a model for the segmentation of cursive handwriting into strokes that has been derived in analogy with those proposed in the literature for early processing tasks in primate visual system. The model allows reformulating the problem of selecting on the ink the points corresponding to perceptually relevant changes of curvature as a preattentive, purely bottom-up visual task, where the conspicuity of curvature changes is measured in terms of their saliency. The modeling of the segmentation as a saliency-driven visual task has lead to a segmentation algorithm whose architecture is biologically-plausible and that does not rely on any parameter other than those that can be directly obtained from the ink. Experimental results show that the performance is very stable and predictable, thus preventing those erratic behaviors of segmentation methods often reported in the literature. They also suggest that the proposed measure of saliency has a direct relation with the dynamics of the handwriting, so as it could be used to capture in a quantitative way some aspects of cursive handwriting intuitively related to the notion of style.

In the framework of online cursive handwriting recognition, we present an efficient method for reordering the sequence of strokes composing handwriting in two special cases of interest: the horizontal bar of the character "" and the dot of the character "". The proposed method exploits shape information for selecting the strokes that most likely correspond to the features of interest, and layout and topological information for locating the strokes representing the body of the characters to which the features belong to. The method does not depend on the specific algorithm used for detecting the elementary strokes in which the electronic ink may be decomposed into. The performance of our method, evaluated on a data set of cursive words produced by 50 different writers, has shown a correct reordering of the sequence in more than 85% of the cases. Thus, the proposed method allows obtaining a more stable and invariant description of the electronic ink in terms of elementary stroke sequences, and therefore can be helpfully used as a preprocessing step for both segmentation-based and word-based handwriting recognition systems.

We propose a characterization of the dynamic behavior of an evolutionary algorithm (EA) with fitness sharing as a function of both the niche radius and the population size. Such a characterization, given in terms of the mean and the standard deviation of the number of niches found during the evolution, can be applied to any EA employing a proportional selection mechanism and does not make any assumption on either the fitness landscape or the internal parameters of the EA itself. On the basis of the proposed characterization, a method for estimating the optimal values for the population size and the niche radius without any a priori information on the fitness landscape is presented and tested on a standard set of functions. The proposed method also provides the best solution for the problem at hand, i.e., the solution obtained in correspondence of such optimal values, at no additional cost.

We introduce a multiple classifier systemthat incorporates a global optimization technique based ona Genetic Algorithm for dynamically selecting the set ofexperts to use in the majority vote approach. The proposedtechnique is applicable when the experts in the pool provideboth the class assigned to the input sample and a measureof the reliability of the this classification. For each sample,the experts selected for participating in the majority voteare those whose reliability is larger than a given threshold.There are as many thresholds as the number of experts bythe number of classes. The values of the thresholds aimedat selecting the best set of experts for each input sampleare determined by a canonical Genetic Algorithm. Thereliability measures provided by the experts of the pool arealso used to implement the tie-break mechanism neededwithin the majority vote scheme. The system has beentested on a handwritten digit recognition problem, and itsperformance compared with those exhibited by other multi-expertsystems exploiting different combining rules.

The recent research efforts in the field of video parsing and analysis have recognized that the soundtrack represents an important supplementary source of content information. In this framework, one of the most relevant topics is that of detecting homogeneous segments within the audio stream, in that changes in the audio very often coincide with scene changes. We present some preliminary results obtained by using different evolutionary algorithms for detecting music and speech audio segments. The experiments have been carried out on MPEG encoded sequences to avoid the computational cost of the decoding procedures.

A method defining a reject option that is applicable to a given 0-reject classifier is proposed. The reject option is based on an estimate of the classification reliability, measured by a reliability evaluator /spl Psi/. Trivially, once a reject threshold /spl sigma/ has been fixed, a sample is rejected if the corresponding value of /spl Psi/ is below /spl sigma/. Obviously, as /spl sigma/ represents the least tolerable classification reliability level, when its value varies the reject option becomes more or less severe. In order to adapt the behavior of the reject option to the requirements of the considered application domain, a function P characterizing the reject option's adequacy to the domain has been introduced. It is shown that P can be expressed as a function of /spl sigma/ and, consequently, the optimal value for /spl sigma/ is defined as the one which maximizes the function P. The method for determining the optimal threshold value is independent of the specific 0-reject classifier, while the definition of the reliability evaluators is related to the classifier's architecture. General criteria for defining appropriate reliability evaluators within a classification paradigm are illustrated in the paper and are based on the localization, in the feature space, of the samples that could be classified with a low reliability. The definition of the reliability evaluators for three popular architectures of neural networks (backpropagation, learning vector quantization and probabilistic network) is presented. Finally, the method has been tested with reference to a complex classification problem with data generated according to a distribution-of-distributions model.

We present a handwritten numeral recognition system centered on a novel method for extracting the set of prototypes to be used during the classification. The method is based on an evolutionary learning mechanism that exploits a genetic algorithm with niching for producing the best set of prototypes. By combining the search power of genetic algorithms and the ability of niching mechanisms to maintain different prototypes during the evolution, the proposed method allows to obtain as many prototypes as needed to model the variability exhibited by the samples belonging to each class. Such a learning mechanism overcomes the limitations of other evolutionary learning methods proposed in the literature for dealing with problems characterized by a large amount of variability in the data set as in the case of handwriting recognition. Experiments have proved that the performance of the system is comparable with, or even better than that exhibited by a neural classifier.

This paper proposes a structural description scheme using circular arcs as primitives. On this scheme, a metric for defining a distance between pairs of circular arcs and relations among them, is introduced and its main properties are discussed. This metric is based on a set of perceptive criteria which allow to increase its effectiveness in application domains characterized by high variability in the shape of the visual patterns. The whole approach is general enough to be satisfactorily used in a wide class of applications. The metric has been validated by employing it in a nearest neighbour classifier, which has been used for automatic recognition of handwritten digits extracted from a standard character database.

Criteria for evaluating the classification reliability of a neural classifier and for accordingly making a reject option are proposed. Such an option, implemented by means of two rules which can be applied independently of topology, size, and training algorithms of the neural classifier, allows one to improve the classification reliability. It is assumed that a performance function P is defined which, taking into account the requirements of the particular application, evaluates the quality of the classification in terms of recognition, misclassification, and reject rates. Under this assumption the optimal reject threshold value, determining the best trade-off between reject rate and misclassification rate, is the one for which the function P reaches its absolute maximum. No constraints are imposed on the form of P, but the ones necessary in order that P actually measures the quality of the classification process. The reject threshold is evaluated on the basis of some statistical distributions characterizing the behavior of the classifier when operating without reject option; these distributions are computed once the training phase of the net has been completed. The method has been tested with a neural classifier devised for handprinted and multifont printed characters, by using a database of about 300000 samples. Experimental results are discussed.

It is commonly presumed that only parallel thinning algorithms can be efficiently implemented on a parallel machine. In this paper it is shown that also a sequential thinning algorithm can have parallel features which can be made explicit and successfully used for a parallel implementation. To this end, the main phases of a fully sequential algorithm are reformulated in such a way that each phase can be carried out by using parallel operators. Experimental results, obtained on a general purpose SIMD machine, are finally discussed.

The method presented allows the drawing of templates that are reliable in locating distorted occurrences of the symbol to recognize and robust against the noise and the false alarms present on the background. A learning by showing technique is used to draw the template by considering samples of the symbol coming from a training set. The design of the template involves the evaluation of the reliablity of each pixel, obtained by estimating the entropy which characterizes the pixel. To be considered reliable, a pixel must show an entropy less than a threshold H/sub o/ set so as to maximize a match performance figure: the obtained template is constituted only by the pixels meeting this requirement.>

A neural classifier for isolated omnifont characters is discussed. A method for characterizing a given training set of characters, based on the definition of some statistical parameters is introduced; on the basis of such characterization an architecture is defined made of a set of neural networks properly connected. Depending on the value of the parameters characterizing the training set, both sizing and training of each network are separately carried out according to a suitable methodology. It is shown that higher recognition rates can be achieved than those obtained by using a single neural network as classifier.>

Presents a method for the recognition of symbols on binary images, especially tailored for geographic maps. The prototyping stage is performed by means of a geometric approach: using a suitable distance definition, the prototype of a class is obtained as the pattern minimizing the sum of distances among itself and all the samples of the considered class. In this phase some statistical parameters characterizing the distortions occurring on the symbols are also obtained, so allowing the estimate of the recognition reliability. On the basis of the misclassification and reject costs, some tuning parameters, affecting the classification criteria, are also evaluated in order to maximize the classification performances. Experimental results for several test maps are finally presented.>