Showing papers on "Feature vector published in 1994"

Fast subsequence matching in time-series databases

Abstract: We present an efficient indexing method to locate 1-dimensional subsequences within a collection of sequences, such that the subsequences match a given (query) pattern within a specified tolerance. The idea is to map each data sequences into a small set of multidimensional rectangles in feature space. Then, these rectangles can be readily indexed using traditional spatial access methods, like the R*-tree [9]. In more detail, we use a sliding window over the data sequence and extract its features; the result is a trail in feature space. We propose an efficient and effective algorithm to divide such trails into sub-trails, which are subsequently represented by their Minimum Bounding Rectangles (MBRs). We also examine queries of varying lengths, and we show how to handle each case efficiently. We implemented our method and carried out experiments on synthetic and real data (stock price movements). We compared the method to sequential scanning, which is the only obvious competitor. The results were excellent: our method accelerated the search time from 3 times up to 100 times.

Efficient and effective querying by image content

Abstract: In the QBIC (Query By Image Content) project we are studying methods to query large on-line image databases using the images' content as the basis of the queries. Examples of the content we use include color, texture, shape, position, and dominant edges of image objects and regions. Potential applications include medical (“Give me other images that contain a tumor with a texture like this one”), photo-journalism (“Give me images that have blue at the top and red at the bottom”), and many others in art, fashion, cataloging, retailing, and industry. We describe a set of novel features and similarity measures allowing query by image content, together with the QBIC system we implemented. We demonstrate the effectiveness of our system with normalized precision and recall experiments on test databases containing over 1000 images and 1000 objects populated from commercially available photo clip art images, and of images of airplane silhouettes. We also present new methods for efficient processing of QBIC queries that consist of filtering and indexing steps. We specifically address two problems: (a) non Euclidean distance measures; and (b) the high dimensionality of feature vectors. For the first problem, we introduce a new theorem that makes efficient filtering possible by bounding the non-Euclidean, full cross-term quadratic distance expression with a simple Euclidean distance. For the second, we illustrate how orthogonal transforms, such as Karhunen Loeve, can help reduce the dimensionality of the search space. Our methods are general and allow some “false hits” but no false dismissals. The resulting QBIC system offers effective retrieval using image content, and for large image databases significant speedup over straightforward indexing alternatives. The system is implemented in X/Motif and C running on an RS/6000.

The TV-tree: an index structure for high-dimensional data

Abstract: We propose a file structure to index high-dimensionality data, which are typically points in some feature space. The idea is to use only a few of the features, using additional features only when the additional discriminatory power is absolutely necessary. We present in detail the design of our tree structure and the associated algorithms that handle such "varying length" feature vectors. Finally, we report simulation results, comparing the proposed structure with the R*-tree, which is one of the most successful methods for low-dimensionality spaces. The results illustrate the superiority of our method, which saves up to 80% in disk accesses.

A moment-based approach to registration of images with affine geometric distortion

Abstract: A new method is described for automatic control point selection and matching. First, reference and sensed images are segmented and closed-boundary regions are extracted. Each region is represented by a set of affine-invariant moment-based features. Correspondence between the regions is then established by a two-stage matching algorithm that works both in the feature space and in the image space. Centers of gravity of corresponding regions are used as control points. A practical use of the proposed method is demonstrated by registration of SPOT and Landsat TM images. It is shown that the authors' method can produce subpixel registration accuracy. >

Handwriting signal processing front-end for handwriting recognizers

Abstract: A handwriting signal processing front-end method and apparatus for a handwriting training and recognition system which includes non-uniform segmentation and feature extraction in combination with multiple vector quantization. In a training phase, digitized handwriting samples are partitioned into segments of unequal length. Features are extracted from the segments and are grouped to form feature vectors for each segment. Groups of adjacent from feature vectors are then combined to form input frames. Feature-specific vectors are formed by grouping features of the same type from each of the feature vectors within a frame. Multiple vector quantization is then performed on each feature-specific vector to statistically model the distributions of the vectors for each feature by identifying clusters of the vectors and determining the mean locations of the vectors in the clusters. Each mean location is represented by a codebook symbol and this information is stored in a codebook for each feature. These codebooks are then used to train a recognition system. In the testing phase, where the recognition system is to identify handwriting, digitized test handwriting is first processed as in the training phase to generate feature-specific vectors from input frames. Multiple vector quantization is then performed on each feature-specific vector to represent the feature-specific vector using the codebook symbols that were generated for that feature during training. The resulting series of codebook symbols effects a reduced representation of the sampled handwriting data and is used for subsequent handwriting recognition.

Voxel similarity measures for automated image registration

Abstract: We present the concept of the feature space sequence: 2D distributions of voxel features of two images generated at registration and a sequence of misregistrations. We provide an explanation of the structure seen in these images. Feature space sequences have been generated for a pair of MR image volumes identical apart from the addition of Gaussian noise to one, MR image volumes with and without Gadolinium enhancement, MR and PET-FDG image volumes and MR and CT image volumes, all of the head. The structure seen in the feature space sequences was used to devise two new measures of similarity which in turn were used to produce plots of cost versus misregistration for the 6 degrees of freedom of rigid body motion. One of these, the third order moment of the feature space histogram, was used to register the MR image volumes with and without Gadolinium enhancement. These techniques have the potential for registration accuracy to within a small fraction of a voxel or resolution element and therefore interpolation errors in image transformation can be the dominant source of error in subtracted images. We present a method for removing these errors using sinc interpolation and show how interpolation errors can be reduced by over two orders of magnitude.

Automatic method for scoring and clustering prototypes of handwritten stroke-based data

Abstract: A system and method for processing stroke-based handwriting data for the purposes of automatically scoring and clustering the handwritten data to form letter prototypes. The present invention includes a method for processing digitized stroke-based handwriting data of known character strings, where each of the character strings is represented by a plurality of mathematical feature vectors. In this method, each one of the plurality of feature vectors is labelled as corresponding to a particular character in the character strings. A trajectory is then formed for each one of the plurality of feature vectors labelled as corresponding to a particular character. After the trajectories are formed, a distance value is calculated for each pair of trajectories corresponding to the particular character using dynamic time warping method. The trajectories which are within a sufficiently small distance of each other are grouped to form a plurality of clusters. The clusters are used to define handwriting prototypes which identify subcategories of the character.

Method and apparatus for image recognition using invariant feature signals

Abstract: A method of operating an image recognition system including providing a neural network including a plurality of input neurons, a plurality of output neurons and an interconnection weight matrix; providing a display including an indicator; initializing the indicator to an initialized state; obtaining an image of a structure; digitizing the image so as to obtain a plurality of input intensity cells and define an input object space; transforming the input object space to a feature vector including a set of n scale-, position- and rotation- invariant feature signals, where n is a positive integer not greater than the plurality of input neurons, by extracting the set of n scale-, position- and rotation-invariant feature signals from the input object space according to a set of relationships I k =∫.sub.Ω ∫I(x,y)h[k,I(x,y)]dxdy, where I k is the set of n scale-, position- and rotation-invariant feature signals, k is a series of counting numbers from 1 to n inclusive, (x,y) are the coordinates of a given cell of the plurality of input intensity cells, I(x,y) is a function of an intensity of the given cell of the plurality of input intensity cells, Ω is an area of integration of input intensity cells, and h[k,I(x,y)] is a data dependent kernel transform from a set of orthogonal functions, of I(x,y) and k; transmitting the set of n scale-, position- and rotation- invariant feature signals to the plurality of input neurons; transforming the set of n scale-, position- and rotation- invariant feature signals at the plurality of input neurons to a set of structure recognition output signals at the plurality of output neurons according to a set of relationships defined at least in part by the interconnection weight matrix of the neural network; transforming the set of structure recognition output signals to a structure classification signal; and transmitting the structure classification signal to the display so as to perceptively alter the initialized state of the indicator and display the structure recognition signal for the structure.

Gray-scale morphological granulometric texture classification

Abstract: Binary morphological granulometric size distributions were conceived by Matheron as a way of describing image granularity (or texture). Since each normalized size distribution is a probability density, feature vectors of granulometric moments result. Recent application has focused on taking local size distributions around individual pixels so that the latter can be classified by surrounding texture. The extension of the local-classification technique to gray-scale textures is investigated. It does so by using 42 granulometric features, half generated by opening granulometries and a dual half generated by closing granulometries. After training and classification of both dependent and independent data, feature extraction (compression) is accomplished by means of the Karhunen-Loeve transform. Sequential feature selection is also applied. The effect of randomly placed uniform noise is investigated. In particular, the degree to which training in noise increases robustness across noise levels is studied, and feature selection is employed to arrive at a noise-insensitive set of granulometric classifiers.

On-line cursive script recognition using time delay neural networks and hidden Markov models

Abstract: Presents a writer independent system for on-line handwriting recognition which can handle both cursive script and hand-print. The pen trajectory is recorded by a touch sensitive pad, such as those used by note-pad computers. The input to the system contains the pen trajectory information, encoded as a time-ordered sequence of feature vectors. Features include X and Y coordinates, pen-lifts, speed, direction and curvature of the pen trajectory. A time delay neural network with local connections and shared weights is used to estimate a posteriori probabilities for characters in a word. A hidden Markov model segments the word into characters in a way which optimizes the global word score, taking a dictionary into account. A geometrical normalization scheme and a fast but efficient dictionary search are also presented. Trained on 20000 unconstrained cursive words from 59 writers and using a 25000 word dictionary the authors reached a 89% character and 80% word recognition rate on test data from a disjoint set of writers. >

Learning and feature selection in stereo matching

Abstract: We present a novel stereo matching algorithm which integrates learning, feature selection, and surface reconstruction. First, a new instance based learning (IBL) algorithm is used to generate an approximation to the optimal feature set for matching. In addition, the importance of two separate kinds of knowledge, image dependent knowledge and image independent knowledge, is discussed. Second, we develop an adaptive method for refining the feature set. This adaptive method analyzes the feature error to locate areas of the image that would lead to false matches. Then these areas are used to guide the search through feature space towards maximizing the class separation distance between the correct match and the false matches. Third, we introduce a self-diagnostic method for determining when apriori knowledge is necessary for finding the correct match. If the a priori knowledge is necessary then we use a surface reconstruction model to discriminate between match possibilities. Our algorithm is comprehensively tested against fixed feature set algorithms and against a traditional pyramid algorithm. Finally, we present and discuss extensive empirical results of our algorithm based on a large set of real images. >

New results on traffic sign recognition

Abstract: In cooperation with Daimler-Benz AG within the European PROMETHEUS project the authors are developing a real-time traffic sign recognition system. The system is installed in a test vehicle of Daimler-Benz that operates autonomously on European highways and performs a vision based environmental analysis that includes, among others, road tracking, obstacle detection, lane traffic analysis and traffic sign recognition. Within the traffic sign recognition the authors' group provides a real-time detection and classification of traffic signs. The authors' traffic sign classifier (TSC) is based on the color segmentation system CSC (Color Structure Code) developed in their group. The authors present details of their latest CSC-evaluation strategies for the TSC. The authors have succeeded in developing a feature space for the localization of all traffic sign candidates. The main features of an object are the typical color, its form and the inclusion of certain forms and colors. The feature space allows a fast classification of the traffic sign candidates following a hierarchical decision graph. The overall decision of whether it is a traffic sign or not is made by a fuzzy control system handling the probabilities of all the single decisions. In addition the authors present several statistical evaluations for some thousands of images that will show the quality of their TSC. For a real-time behavior the authors tested different hardware components (C40, Motorola PC601, T 805) in a TIP system (transputer image processing) and they explain the results.

Temporal texture and activity recognition

Abstract: The goal of this thesis is to demonstrate the utility of low-level motion features for the purpose of recognition. Although motion plays an important role in biological recognition tasks, motion recognition, in general, has received little attention in the literature compared to the volume of work on static object recognition. It has been shown that in some cases, motion information alone is sufficient for human visual system to achieve reliable recognition. Previous attempts at duplicating such capability in machine vision have been based on abstract higher-level models of objects, or have required building intermediate representations such as the trajectories of certain feature points of the object. In this work we demonstrate that motion recognition can be accomplished using lower-level motion features, without the use of abstract object models or trajectory representations. First, we show that certain statistical spatial and temporal features derived from the optic flow field have invariant properties, and can be used to classify regional motion patterns such as ripples on water, fluttering of leaves, and chaotic fluid flow. We then present a novel low-level computational approach for detecting and recognizing temporally repetitive movements, such as those characteristic of walking people, or flying birds, on the basis of the periodic nature of their motion signatures. We demonstrate the techniques on a number of real-world image sequences containing complex non-rigid motion patterns. We also show that the proposed techniques are reliable and efficient by implementing a real-time activity recognition system.

Detection of human speech in structured noise

Abstract: This paper describes research to develop an efficient system that provides a binary decision as to the presence of speech in a short (one to three second) time sample of an acoustic signal. A method which is efficient and reliably detects human speech in the presence of structured noise (such as wind, music, traffic sounds, etc.) is described. Two separate algorithms were developed. The first algorithm detects the presence of speech by testing for concave and/or convex formant shapes. The second algorithm is a statistical pattern classifier utilizing radial basis function (RBF) networks with mel-cepstra feature vectors. Classification errors are not consistent across these two different methods. As a consequence, we plan to reduce our error rate by fusion of these methods. >

MECA: maximum entropy clustering algorithm

Abstract: This paper presents a new approach to fuzzy clustering, which provides the basis for the development of the maximum entropy clustering algorithm (MECA). The derivation of the proposed algorithm is based on an objective function incorporating a measure of the entropy of the membership functions and a measure of the distortion between the prototypes and the feature vectors. This formulation allows the gradual transition from a maximum uncertainty or minimum selectivity phase to a minimum uncertainty or maximum selectivity phase during the clustering process. Such a transition is achieved by controlling the relative effect of the maximization of the membership entropy and the minimization of the distortion between the prototypes and the feature vectors. The IRIS data set provides the basis for evaluating the proposed algorithms and comparing their performance with that of competing techniques. >

Efficient Methods for Dealing with Missing Data in Supervised Learning

Abstract: We present efficient algorithms for dealing with the problem of missing inputs (incomplete feature vectors) during training and recall. Our approach is based on the approximation of the input data distribution using Parzen windows. For recall, we obtain closed form solutions for arbitrary feedforward networks. For training, we show how the backpropagation step for an incomplete pattern can be approximated by a weighted averaged backpropagation step. The complexity of the solutions for training and recall is independent of the number of missing features. We verify our theoretical results using one classification and one regression problem.

Method and apparatus of pattern recognition

Abstract: A pattern recognition apparatus and method in which a set is produced which includes a fundamental pattern vector on a basis place and other fundamental pattern vectors of the patterns displaced from the fundamental pattern on the basis place. Then a subspace spanned by fundamental pattern vectors included in the set is generated. A test pattern vector of a wafer to be inspected is projected to the subspace and similarity between the fundamental vectors and the test pattern vector is measured. Further, an image is used after it is filtered by a normalization filter. Furthermore, sensitivity of pattern recognition is varied by changing the dimension of the pattern vectors. Moreover, for objects expressed by numerical values which can not be compared directly, the data of the objects are transformed into images and, then, a set of fundamental pattern vectors are worked out.

A neural network approach to classification of sidescan sonar imagery from a midocean ridge area

Abstract: A neural-network approach to classification of sidescan-sonar imagery is tested on data from three distinct geoacoustic provinces of a midocean-ridge spreading center: axial valley, ridge flank, and sediment pond. The extraction of representative features from the sidescan imagery is analyzed, and the performance of several commonly used texture measures are compared in terms of classification accuracy using a backpropagation neural network. A suite of experiments compares the effectiveness of different feature vectors, the selection of training patterns, the configuration of the neural network, and two widely used statistical methods: Fisher-pairwise classifier and nearest-mean algorithm with Mahalanobis distance measure. The feature vectors compared here comprise spectral estimates, gray-level run length, spatial gray-level dependence matrix, and gray-level differences. The overall accurate classification rates using the best feature set for the three seafloor types are: sediment ponds, 85.9%; ridge flanks, 91.2%; and valleys, 80.1%. While most current approaches are statistical, the significant finding in this study is that high performance for seafloor classification in terms of accuracy and computation can be achieved using a neural network with the proper combination of texture features. These are preliminary results of our program toward the automated segmentation and classification of undersea terrain. >

NNERVE: Neural Network Extraction of Repetitive Vectors for Electromyography. I. Algorithm

Abstract: Artificial neural network (ANN) based signal processing methods have been shown to have significant robustness in processing complex, degraded, noisy, and unstable signals. A novel approach to automated electromyogram (EMG) signal decomposition, using an ANN processing architecture, is presented here. Due to the lack of a priori knowledge of motor unit action potential (MUAP) morphology, the EMG decomposition must be performed in an unsupervised manner. An ANN classifier, consisting of a multilayer perceptron neural network and employing a novel unsupervised training strategy, is proposed. The ANN learns repetitive appearances of MUAP waveforms from their suspected occurrences in a filtered EMG signal in an autoassociative learning task. The same training waveforms are fed into the trained ANN and the output of the ANN is fed back to its input, giving rise to a dynamic retrieval net classifier. For each waveform in the data, the network discovers a feature vector associated with that waveform. For each waveform, classification is achieved by comparing its feature vector with those of the other waveforms. Firing information of each MUAP is further used to refine the classification results of the ANN classifier. Then, individual MUAP waveform shapes are derived and their firing tables are created. >

Multiscale image texture analysis in wavelet spaces

Abstract: The paper describes a new method for texture feature extraction and analysis in images using wavelet transform (WT), KL-expansion and Kohonen maps. For this purpose, the authors first apply a global wavelet transform on the initial image. Due to the localization properties of the WT both in the spatial and in the frequency domain it is possible to describe the local texture features in the surroundings of any pixel by a set of respective wavelet coefficients. This is accomplished by a local traversal of the wavelet pyramid and finally results in the feature vector required. Since the localization is limited by Heisenberg's uncertainty principle one must approximate the single coefficients for each pixel by piecewise linear interpolation. Once the feature vector is derived from the WT, further steps in the analysis pipeline perform decorrelation, normalization and finally clustering and supervised classification. In contrast to many related wavelet-based approaches, that usually apply different WTs on every texture sample and classify based on means derived from the former, the present method especially accounts for many real world applications. In those cases there are not usually large coherent texture regions that allow separated treatment. Moreover the approach employs a global WT and then stresses the local properties of the basis functions to identify local areas of interest from the initial image, as for instance training areas. The authors illustrate the efficiency of the method by classifying different real world textures with LVQ classifiers. >

An invariant traffic sign recognition system based on sequential color processing and geometrical transformation

Abstract: One of the most noteworthy problems associated with conventional pattern recognition methods is that it is not easy to extract feature vectors from images which are not translation, rotation, and scale change invariant in outdoor noisy environments. This paper describes the development of an invariant traffic sign recognition system capable of tolerating the above variations. The signs are restricted to three types of warning signs and are all of red color. The developed method is insensitive to brightness changes as well as invariant to translation, rotation, scale change, and noise. The architecture of this system is based upon neural network supervised learning after geometrical transformations have been applied. The performance of this system is compared with other invariant approaches in terms of the percentage of correct decisions in outdoor noisy environments. >

Method and apparatus for inferring the topical content of a document based upon its lexical content without supervision

Abstract: An iterative method of determining the topical content of a document using a computer. The processing unit of the computer determines the topical content of documents presented to it in machine readable form using information stored in computer memory. That information includes word-clusters, a lexicon, and association strength values. The processing unit beings by generating an observed feature vector for the document being characterized, which indicates which of the words of the lexicon appear in the document. Afterward, the processing unit makes an initial prediction of the topical content of the document in the form of a topic belief vector. The processing unit uses the topic belief vector and the association strength values to predict which words of the lexicon should appear in the document. This prediction is represented via a predicted feature vector. The predicted feature vector is then compared to the observed feature vector to measure how well the topic belief vector models the topical content of the document. If the topic belief vector adequately model the topical content of the document, then the processing unit's task is complete. On the other hand, if the topic belief vector does not adequately model the topical content of the document, then the processing unit determines how the topic belief vector should be modified to improve the prediction of modeling of the topical content.

Analysis of phoneme-based features for language identification

Abstract: This paper presents an analysis of the phonemic language identification system introduced previously (see Eurospeech, vol.2, p.1307, 1993), now extended to recognize German in addition to English and Japanese. In this system language identification is based on features derived from a superset of phonemes of all three languages. As we increase the number of languages, the need to reduce the feature space becomes apparent. Practical analysis of single-feature statistics in conjunction with linguistic knowledge leads to 90% reduction of the feature space with only a 5% loss in performance. Thus, the system discriminates between Japanese and English with 84.1% accuracy based on only 15 features compared to 84.6% based on the complete set of 318 phonemic features (or 83.6% using 333 broad-category features). Results indicate that a language identification system may be designed based on linguistic knowledge and then implemented with a neural network of appropriate complexity. >

Method of generating dictionary for pattern recognition and pattern recognition method using the same

Abstract: A reference pattern vector is obtained from training patterns belonging to each class and is held as a parameter of an original distance function in a distance dictionary. A discriminant function is defined which has, as variables, the difference between respective corresponding components of a feature vector of each training pattern and the corresponding reference pattern vector and the square of the difference. Training patterns of all classes are discriminated with the original distance function and a rival pattern set, which includes patterns misclassified as belonging to a respective class, is derived from the results of discrimination of the training patterns. A discriminant analysis is made between the training pattern set of each class and the corresponding rival pattern set to thereby determine parameters of the discriminant function, which are held in a discriminant dictionary. The original distance function and the discriminant function are additively coupled together by a predetermined coupling coefficient to define a learned distance function, which is used to discriminate the training patterns to update the learned distance function.

CANDID: comparison algorithm for navigating digital image databases

Abstract: In this paper, we propose a method for calculating the similarity between two digital images. A global signature describing the texture, shape, or color content is first computed for every image stored in a database, and a normalized distance between probability density functions of feature vectors is used to match signatures. This method can be used to retrieve images from a database that are similar to an example target image. This algorithm is applied to the problem of search and retrieval for a database containing pulmonary CT imagery, and experimental results are provided. >

Transient sonar signal classification using hidden Markov models and neural nets

Abstract: In ocean surveillance, a number of different types of transient signals are observed. These sonar signals are waveforms in one dimension (1-D). The hidden Markov model (HMM) is well suited to classification of 1-D signals such as speech. In HMM methodology, the signal is divided into a sequence of frames, and each frame is represented by a feature vector. This sequence of feature vectors is then modeled by one HMM. Thus, the HMM methodology is highly suitable for classifying the patterns that are made of concatenated sequences of micro patterns. The sonar transient signals often display an evolutionary pattern over the time scale. Following this intuition, the application of HMM's to sonar transient classification is proposed and discussed in this paper. Toward this goal, three different feature vectors based on an autoregressive (AR) model, Fourier power spectra, and wavelet transforms are considered in our work. In our implementation, one HMM is developed for each class of signals. During testing, the signal to be recognized is matched against all models. The best matched model identifies the signal class. The neural net (NN) classifier has been successfully used previously for sonar transient classification. The same set of features as mentioned above is then used with a multilayer perceptron NN classifier. Some experimental results using "DARPA standard data set I" with HMM and MLP-NN classification schemes are presented. A combined NN/HMM classifier is proposed, and its performance is evaluated with respect to individual classifiers. >

Finite-element method for image alignment and morphing

Abstract: In a morphing system for creating intermediate images that, viewed serially, make an object in a source image appear to metamorphose into a different object in a target image, correspondences between feature points in the source and target images are determined by simulating the modes of motion of respective elastic sheets in which are embedded nodal points that correspond to respective feature points in the images. For each feature point, a generalized-feature vector is determined whose components represent the associated nodal point's participations in the various motion modes. Correspondences between feature points in the source and target images are determined in accordance with the closeness of the points' generalized feature vectors. Correspondences thus determined can additionally be used for alignment and object-recognition purposes.

Classification of cervical cell nuclei using morphological segmentation and textural feature extraction

Abstract: This paper presents preliminary results for the classification of Pap Smear cell nuclei, using gray level co-occurrence matrix (GLCM) textural features. We outline a method of nuclear segmentation using fast morphological gray-scale transforms. For each segmented nucleus, features derived from a modified form of the GLCM are extracted over several angle and distance measures. Linear discriminant analysis is performed on these features to reduce the dimensionality of the feature space, and a classifier with hyper-quadric decision surface is implemented to classify a small set of normal and abnormal cell nuclei. Using 2 features, we achieve a misclassification rate of 3.3% on a data set of 61 cells. >

An extended-shadow-code based approach for off-line signature verification. I. Evaluation of the bar mask definition

Abstract: In this paper, the authors present an evaluation of the extended shadow code (ESC) used as a global feature vector for the signature verification problem. The proposed class of shape factors seems to be a good compromise between global features related to the general aspect of the signature, and local features related to measurements taken on specific parts of the signature. This is achieved by the bar mask definition, where at low resolution the ESC is related to the overall proportions of the signature. At high resolution, values of the horizontal, vertical and diagonal bars could be related to local measurements taken on specific parts of the signature without requiring low-level handwriting segmentation which is a very difficult task.