Showing papers in "Pattern Recognition in 2002"
TL;DR: The various applications of neural networks in image processing are categorised into a novel two-dimensional taxonomy for image processing algorithms and their specific conditions are discussed in detail.
Abstract: We review more than 200 applications of neural networks in image processing and discuss the present and possible future role of neural networks, especially feed-forward neural networks, Kohonen feature maps and Hopfield neural networks. The various applications are categorised into a novel two-dimensional taxonomy for image processing algorithms. One dimension specifies the type of task performed by the algorithm: preprocessing, data reduction/feature extraction, segmentation, object recognition, image understanding and optimisation. The other dimension captures the abstraction level of the input data processed by the algorithm: pixel-level, local feature-level, structure-level, object-level, object-set-level and scene characterisation. Each of the six types of tasks poses specific constraints to a neural-based approach. These specific conditions are discussed in detail. A synthesis is made of unresolved problems related to the application of pattern recognition techniques in image processing and specifically to the application of neural networks. Finally, we present an outlook into the future application of neural networks and relate them to novel developments.
1,100 citations
TL;DR: Experiments on a database containing a total of 1232 signatures of 102 individuals show that writer-dependent thresholds yield better results than using a common threshold.
Abstract: We describe a method for on-line handwritten signature verification. The signatures are acquired using a digitizing tablet which captures both dynamic and spatial information of the writing. After preprocessing the signature, several features are extracted. The authenticity of a writer is determined by comparing an input signature to a stored reference set (template) consisting of three signatures. The similarity between an input signature and the reference set is computed using string matching and the similarity value is compared to a threshold. Several approaches for obtaining the optimal threshold value from the reference set are investigated. The best result yields a false reject rate of 2.8% and a false accept rate of 1.6%. Experiments on a database containing a total of 1232 signatures of 102 individuals show that writer-dependent thresholds yield better results than using a common threshold.
595 citations
TL;DR: A new method to detect and count bright spots in fluorescence images coming from biological immunomicroscopy experiments is presented, based on the multiscale product of subband images resulting from the a trous wavelet transform decomposition of the original image, after thresholding of non-significant coefficients.
Abstract: We present a new method to detect and count bright spots in fluorescence images coming from biological immunomicroscopy experiments. It is based on the multiscale product of subband images resulting from the a trous wavelet transform decomposition of the original image, after thresholding of non-significant coefficients. The multiscale correlation of the filtered wavelet coefficients, which allows to enhance multiscale peaks due to spots while reducing noise, combines information coming from different levels of resolution and gives a clear and distinctive chacterization of the spots. Results are presented for the analysis of typical immunofluorescence images.
552 citations
TL;DR: This article presents a detailed review of some of the most used calibrating techniques in which the principal idea has been to present them all with the same notation.
Abstract: Camera calibrating is a crucial problem for further metric scene measurement. Many techniques and some studies concerning calibration have been presented in the last few years. However, it is still difficult to go into details of a determined calibrating technique and compare its accuracy with respect to other methods. Principally, this problem emerges from the lack of a standardized notation and the existence of various methods of accuracy evaluation to choose from. This article presents a detailed review of some of the most used calibrating techniques in which the principal idea has been to present them all with the same notation. Furthermore, the techniques surveyed have been tested and their accuracy evaluated. Comparative results are shown and discussed in the article. Moreover, code and results are available in internet.
536 citations
TL;DR: This AFCM algorithm has successfully been used in segmenting the magnetic resonance image of Ophthalmology to differentiate the abnormal tissues from the normal tissues and is recommended for use in cluster analysis.
Abstract: In this paper we propose a new metric to replace the Euclidean norm in c-means clustering procedures. On the basis of the robust statistic and the influence function, we claim that the proposed new metric is more robust than the Euclidean norm. We then create two new clustering methods called the alternative hard c-means (AHCM) and alternative fuzzy c-means (AFCM) clustering algorithms. These alternative types of c-means clustering have more robustness than c-means clustering. Numerical results show that AHCM has better performance than HCM and AFCM is better than FCM. We recommend AFCM for use in cluster analysis. Recently, this AFCM algorithm has successfully been used in segmenting the magnetic resonance image of Ophthalmology to differentiate the abnormal tissues from the normal tissues.
517 citations
TL;DR: This paper considers invariant texture analysis, and approaches whose performances are not affected by translation, rotation, affine, and perspective transform are addressed.
Abstract: This paper considers invariant texture analysis. Texture analysis approaches whose performances are not affected by translation, rotation, affine, and perspective transform are addressed. Existing invariant texture analysis algorithms are carefully studied and classified into three categories: statistical methods, model based methods, and structural methods. The importance of invariant texture analysis is presented first. Each approach is reviewed according to its classification, and its merits and drawbacks are outlined. The focus of possible future work is also suggested.
478 citations
TL;DR: The searching capability of genetic algorithms has been exploited for automatically evolving the number of clusters as well as proper clustering of any data set and the proposed technique is able to distinguish some characteristic landcover types in the image.
Abstract: In this article the searching capability of genetic algorithms has been exploited for automatically evolving the number of clusters as well as proper clustering of any data set. A new string representation, comprising both real numbers and the do not care symbol, is used in order to encode a variable number of clusters. The Davies–Bouldin index is used as a measure of the validity of the clusters. Effectiveness of the genetic clustering scheme is demonstrated for both artificial and real-life data sets. Utility of the genetic clustering technique is also demonstrated for a satellite image of a part of the city Calcutta. The proposed technique is able to distinguish some characteristic landcover types in the image.
417 citations
TL;DR: The proposed distance measure has the advantage over the traditional distance measures regarding the overlap between two distributions; it takes the similarity of the non-overlapping parts into account as well as that of overlapping parts.
Abstract: A distance measure between two histograms has applications in feature selection, image indexing and retrieval, pattern classification and clustering, etc. We propose a distance between sets of measurement values as a measure of dissimilarity of two histograms. The proposed measure has the advantage over the traditional distance measures regarding the overlap between two distributions; it takes the similarity of the non-overlapping parts into account as well as that of overlapping parts. We consider three versions of the univariate histogram, corresponding to whether the type of measurement is nominal, ordinal, and modulo and their computational time complexities are Θ(b), Θ(b) and O(b2) for each type of measurements, respectively, where b is the number of levels in histograms.
403 citations
TL;DR: Four different fingerprint matching algorithms are combined using the proposed scheme to improve the accuracy of a fingerprint verification system and it is shown that a combination of multiple impressions or multiple fingers improves the verification performance by more than 4% and 5%, respectively.
Abstract: A scheme is proposed for classifier combination at decision level which stresses the importance of classifier selection during combination. The proposed scheme is optimal (in the Neyman–Pearson sense) when sufficient data are available to obtain reasonable estimates of the join densities of classifier outputs. Four different fingerprint matching algorithms are combined using the proposed scheme to improve the accuracy of a fingerprint verification system. Experiments conducted on a large fingerprint database (∼2700 fingerprints) confirm the effectiveness of the proposed integration scheme. An overall matching performance increase of ∼3% is achieved. We further show that a combination of multiple impressions or multiple fingers improves the verification performance by more than 4% and 5%, respectively. Analysis of the results provide some insight into the various decision-level classifier combination strategies.
371 citations
TL;DR: This paper presents a comprehensive survey of the use of these pattern recognition methods which enable image and video retrieval bycontent in the media archives.
Abstract: The need for content-based access to image and video information from media archives has captured the attention of researchers in recent years. Research e0orts have led to the development of methods that provide access to image and video data. These methods have their roots in pattern recognition. The methods are used to determine the similarity in the visual information content extracted from low level features. These features are then clustered for generation of database indices. This paper presents a comprehensive surveyon the use of these pattern recognition methods which enable image and video retrieval bycontent. ? 2002 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.
355 citations
IBM1
TL;DR: Biometrics authentication systems suffer from some inherent biometrics-specific security threats, mainly related to the use of digital signals and the need for additional input devices, though the also discuss brute-force attacks of biometric systems.
Abstract: Biometrics authentication offers many advantages over conventional authentication systems that rely on possessions or special knowledge. With conventional technology, often the mere possession of an employee ID card is proof of ID, while a password potentially can be used by large groups of colleagues for long times without change. The fact that biometrics authentication is non-repudiable (hard to refute) and, yet, convenient, is among its most important advantages. Biometrics systems, however, suffer from some inherent biometrics-specific security threats. These threats are mainly related to the use of digital signals and the need for additional input devices, though we also discuss brute-force attacks of biometrics systems. There are also problems common to any pattern recognition system. These include “wolves” and “lambs”, and a new group we call “chameleons”. An additional issue with the use of biometrics is the invasion of privacy because the user has to enroll with an image of a body part. We discuss these issues and suggest some methods for mitigating their impact.
TL;DR: This paper uses tabu search to solve this feature selection problem and compares it with classic algorithms, such as sequential methods, branch and bound method, etc., and most other suboptimal methods proposed recently.
Abstract: Selecting an optimal subset from original large feature set in the design of pattern classifier is an important and difficult problem. In this paper, we use tabu search to solve this feature selection problem and compare it with classic algorithms, such as sequential methods, branch and bound method, etc., and most other suboptimal methods proposed recently, such as genetic algorithm and sequential forward (backward) floating search methods. Based on the results of experiments, tabu search is shown to be a promising tool for feature selection in respect of the quality of obtained feature subset and computation efficiency. The effects of parameters in tabu search are also analyzed by experiments.
TL;DR: A new method to authenticate individuals based on palmprint identification and verification using a texture-based dynamic selection scheme to facilitate the fast search for the best matching of the sample in the database in a hierarchical fashion is described.
Abstract: Biometric computing offers an effective approach to identify personal identity by using individual's unique, reliable and stable physical or behavioral characteristics. This paper describes a new method to authenticate individuals based on palmprint identification and verification. Firstly, a comparative study of palmprint feature extraction is presented. The concepts of texture feature and interesting points are introduced to define palmprint features. A texture-based dynamic selection scheme is proposed to facilitate the fast search for the best matching of the sample in the database in a hierarchical fashion. The global texture energy, which is characterized with high convergence of inner-palm similarities and good dispersion of inter-palm discrimination, is used to guide the dynamic selection of a small set of similar candidates from the database at coarse level for further processing. An interesting point based image matching is performed on the selected similar patterns at fine level for the final confirmation. The experimental results demonstrate the effectiveness and accuracy of the proposed method.
TL;DR: This survey is divided into two parts, the first one dealing with the general aspects of Cursive Word Recognition, the second one focusing on the applications presented in the literature.
Abstract: This paper presents a survey on off-line Cursive Word Recognition. The approaches to the problem are described in detail. Each step of the process leading from raw data to the final result is analyzed. This survey is divided into two parts, the first one dealing with the general aspects of Cursive Word Recognition, the second one focusing on the applications presented in the literature.
TL;DR: A new method of combining Orthogonal Fourier–Mellin moments (OFMMs) with centroid bounding circle scaling is introduced, which is shown to be useful in characterizing images with large variability.
Abstract: In this paper, we consider the use of orthogonal moments for invariant classification of alphanumeric characters of different size. In addition to the Zernike and pseudo-Zernike moments (ZMs and PZMs) which have been previously proposed for invariant character recognition, a new method of combining Orthogonal Fourier–Mellin moments (OFMMs) with centroid bounding circle scaling is introduced, which is shown to be useful in characterizing images with large variability. Through extensive experimentation using ZMs and OFMMs as features, different scaling methodologies and classifiers, it is shown that OFMMs give the best overall performance in terms of both image reconstruction and classification accuracy.
TL;DR: This paper presents an approach that integrates a potentially powerful fuzzy rule induction algorithm with a rough set-assisted feature reduction method, and the integrated rule generation mechanism maintains the underlying semantics of the feature set.
Abstract: The generation of effective feature pattern-based classification rules is essential to the development of any intelligent classifier which is readily comprehensible to the user. This paper presents an approach that integrates a potentially powerful fuzzy rule induction algorithm with a rough set-assisted feature reduction method. The integrated rule generation mechanism maintains the underlying semantics of the feature set. Through the proposed integration, the original rule induction algorithm (or any other similar technique that generates descriptive fuzzy rules), which is sensitive to the dimensionality of the dataset, becomes usable on classifying patterns composed of a moderately large number of features. The resulting learned ruleset becomes manageable and may outperform rules learned using more features. This, as demonstrated with successful realistic applications, makes the present approach effective in handling real world problems.
TL;DR: It is shown that a state-of-the-art automatic fingerprint verification system can successfully distinguish identical twins though with a slightly lower accuracy than nontwins.
Abstract: Reliable and accurate verification of people is extremely important in a number of business transactions as well as access to privileged information. Automatic verification methods based on physical biometric characteristics such as fingerprint or iris can provide positive verification with a very high accuracy. However, the biometrics-based methods assume that the physical characteristics of an individual (as captured by a sensor) used for verification are sufficiently unique to distinguish one person from another. Identical twins have the closest genetics-based relationship and, therefore, the maximum similarity between fingerprints is expected to be found among identical twins. We show that a state-of-the-art automatic fingerprint verification system can successfully distinguish identical twins though with a slightly lower accuracy than nontwins.
TL;DR: This paper proposes a least-squares solution method using Householder Transformation to find a new representation of face recognition using independent component analysis and demonstrates that not all ICs are useful for recognition.
Abstract: This paper addresses the problem of face recognition using independent component analysis (ICA). More specifically, we are going to address two issues on face representation using ICA. First, as the independent components (ICs) are independent but not orthogonal, images outside a training set cannot be projected into these basis functions directly. In this paper, we propose a least-squares solution method using Householder Transformation to find a new representation. Second, we demonstrate that not all ICs are useful for recognition. Along this direction, we design and develop an IC selection algorithm to find a subset of ICs for recognition. Three public available databases, namely, MIT AI Laboratory, Yale University and Olivette Research Laboratory, are selected to evaluate the performance and the results are encouraging.
TL;DR: In this article, a novel principal component analysis (PCA) technique directly based on original image matrices is developed for image feature extraction, which is more powerful and efficient than conventional PCA and FLD.
Abstract: The conventional principal component analysis (PCA) and Fisher linear discriminant analysis (FLD) are both based on vectors. Rather, in this paper, a novel PCA technique directly based on original image matrices is developed for image feature extraction. Experimental results on ORL face database show that the proposed IMPCA are more powerful and efficient than conventional PCA and FLD.
TL;DR: The experimental results demonstrate that the HOB can find homogeneous regions more effectively than HIB does, and can solve the problem of discriminating shading in color images to some extent.
Abstract: In this paper, a color image segmentation approach based on homogram thresholding and region merging is presented. The homogram considers both the occurrence of the gray levels and the neighboring homogeneity value among pixels. Therefore, it employs both the local and global information. Fuzzy entropy is utilized as a tool to perform homogram analysis for finding all major homogeneous regions at the first stage. Then region merging process is carried out based on color similarity among these regions to avoid oversegmentation. The proposed homogram-based approach (HOB) is compared with the histogram-based approach (HIB). The experimental results demonstrate that the HOB can find homogeneous regions more effectively than HIB does, and can solve the problem of discriminating shading in color images to some extent.
TL;DR: Three procedures, based on curvature coefficient, bi-quadratic interpolation and gradient vector interpolation, are proposed for calculating the curvature of the equi-gray scale curves of an input image to improve the accuracy of handwritten numeral recognition.
Abstract: In this paper, the authors study on the use of gradient and curvature of the gray scale character image to improve the accuracy of handwritten numeral recognition. Three procedures, based on curvature coefficient, bi-quadratic interpolation and gradient vector interpolation, are proposed for calculating the curvature of the equi-gray scale curves of an input image. Then two procedures to compose a feature vector of the gradient and the curvature are described. The efficiency of the feature vectors are tested by recognition experiments for the handwritten numeral database IPTP CDROM1 and NIST SD3 and SD7. The experimental results show the usefulness of the curvature feature and recognition rate of 99.49% and 98.25%, which are one of the highest rates ever reported for these databases (H. Kato et al., Technical Report of IEICE, PRU95-3, 1995, p. 17; R.A. Wilkinson et al., Technical Report NISTIR 4912, August 1992; J. Geist et al., Technical Report NISTIR 5452, June 1994), are achieved, respectively.
TL;DR: This paper compares the strengths and weaknesses of model-based and adaptive techniques, and argues that further advances in the field are likely to involve the increasing use of techniques from the field of artificial intelligence.
Abstract: Research into techniques for the retrieval of images by semantic content is still in its infancy. This paper reviews recent trends in the field, distinguishing four separate lines of activity: automatic scene analysis, model-based and statistical approaches to object classification, and adaptive learning from user feedback. It compares the strengths and weaknesses of model-based and adaptive techniques, and argues that further advances in the field are likely to involve the increasing use of techniques from the field of artificial intelligence.
TL;DR: The problem of where to place the cameras in order to obtain a minimal error in the 3D measurements, also called camera network design in photogrammetry, is dealt with in terms of an optimization design using a multi-cellular genetic algorithm.
Abstract: Three-dimensional (3D) measurements can be recovered from several views by triangulation. This paper deals with the problem of where to place the cameras in order to obtain a minimal error in the 3D measurements, also called camera network design in photogrammetry. We pose the problem in terms of an optimization design, dividing it into two main components: (1) an analytical part dedicated to the analysis of error propagation from which a criterion is derived, and (2) a global optimization process to minimize this criterion. In this way, the approach consists of an uncertainty analysis applied to the reconstruction process from which a covariance matrix is computed. This matrix represents the uncertainty of the detection from which the criterion is derived. Moreover, the optimization has discontinuities due to the presence of occluding surfaces between the viewpoint and the object point group, which leads to a combinatorial optimization process. These aspects are solved using a multi-cellular genetic algorithm. Experimental results are provided to illustrate the effectiveness and efficiency of the solution.
TL;DR: Two e5ective techniques are developed, namely, template condensing and preprocessing, to speed up k-NN classi"cation while maintaining the level of accuracy, and one of the case studies shows that the incorporation of these two techniques to k-nn rule achieves a seven-fold speed-up without sacri"cing accuracy.
Abstract: k-nearest neighbor (k-NN) classi"cation is a well-known decision rule that is widely used in pattern classi"cation. However, the traditional implementation of this method is computationally expensive. In this paper we develop two e5ective techniques, namely, template condensing and preprocessing, to signi"cantly speed up k-NN classi"cation while maintaining the level of accuracy. Our template condensing technique aims at “sparsifying” dense homogeneous clusters of prototypes of any single class. This is implemented by iteratively eliminating patterns which exhibit high attractive capacities. Our preprocessing technique "lters a large portion of prototypes which are unlikely to match against the unknown pattern. This again accelerates the classi"cation procedure considerably, especially in cases where the dimensionality of the feature space is high. One of our case studies shows that the incorporation of these two techniques to k-NN rule achieves a seven-fold speed-up without sacri"cing accuracy. ? 2002 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.
TL;DR: The robustness of the CSS representation under general affine transforms is examined and its performance is compared with the performance of two well-known methods, namely Fourier descriptors and moment invariants.
Abstract: The maxima of curvature scale space (CSS) image have already been used to represent 2-D shapes in different applications. The representation has shown robustness under the similarity transformations. Scaling, orientation changes, translation and even noise can be easily handled by the representation and its associated matching algorithm. In this paper, we examine the robustness of the representation under general affine transforms. We have a database of 1100 images of marine creatures. The contours in this database demonstrate a great range of shape variation. A database of 5000 contours has been constructed using 500 real object boundaries and 4500 contours which are the affine transformed versions of real objects. The CSS representation is then used to find similar shapes from this prototype database. The results provide substantial evidence of stability of the CSS image and its contour maxima under affine transformation. The method is also evaluated objectively through a large classified database and its performance is compared with the performance of two well-known methods, namely Fourier descriptors and moment invariants. The CSS shape descriptor has been selected for MPEG-7 standardization.
TL;DR: This work begins with a filter model, exploiting the geometrical information contained in the minimum spanning tree (MST) built on the learning set, which leads to a feature selection algorithm belonging to a new category of hybrid models ( filter-wrapper).
Abstract: We focus on a hybrid approach of feature selection. We begin our analysis with a filter model , exploiting the geometrical information contained in the minimum spanning tree (MST) built on the learning set. This model exploits a statistical test of relative certainty gain , used in a forward selection algorithm. In the second part of the paper, we show that the MST can be replaced by the 1 nearest-neighbor graph without challenging the statistical framework. This leads to a feature selection algorithm belonging to a new category of hybrid models ( filter-wrapper ). Experimental results on readily available synthetic and natural domains are presented and discussed.
TL;DR: In this article, an optimal 2-D Gabor filter response maximizes a Fisher cost function to discriminate defective texture pixels from non-defective texture pixels, and the results of this optimised Gabor filtering scheme are presented for 35 different flawed homogeneous textures.
Abstract: The task of detecting flaws in woven textiles can be formulated as the problem of segmenting a “known” non-defective texture from an “unknown” defective texture. In order to discriminate defective texture pixels from non-defective texture pixels, optimal 2-D Gabor filters are designed such that, when applied to non-defective texture, the filter response maximises a Fisher cost function. A pixel of potentially flawed texture is classified as defective or non-defective based on the Gabor filter response at that pixel. The results of this optimised Gabor filter classification scheme are presented for 35 different flawed homogeneous textures. These results exhibit accurate flaw detection with low false alarm rate. Potentially, our novel optimised Gabor filter method could be applied to the more complicated problem of detecting flaws in jacquard textiles. This second and more difficult problem is also discussed, along with some preliminary results.
TL;DR: Two new maximum likelihood motion estimation schemes for ultrasound images are presented, based on the assumption that both images are contaminated by a Rayleigh distributed multiplicative noise, which enables motion estimation in cases where a noiseless reference image is not available.
Abstract: When performing block-matching based motion estimation with the ML estimator, one would try to match blocks from the two images, within a predefined search area. The estimated motion vector is that which maximizes a likelihood function, formulated according to the image formation model. Two new maximum likelihood motion estimation schemes for ultrasound images are presented. The new likelihood functions are based on the assumption that both images are contaminated by a Rayleigh distributed multiplicative noise. The new approach enables motion estimation in cases where a noiseless reference image is not available. Experimental results show a motion estimation improvement with regards to other known ML estimation methods.
TL;DR: A two pass algorithm for the segmentation and decomposition of Devanagari composite characters/symbols into their constituent symbols and a recognition rate has been achieved on the segmented conjuncts.
Abstract: Devanagari script is a two dimensional composition of symbols It is highly cumbersome to treat each composite character as a separate atomic symbol because such combinations are very large in number This paper presents a two pass algorithm for the segmentation and decomposition of Devanagari composite characters/symbols into their constituent symbols The proposed algorithm extensively uses structural properties of the script In the first pass, words are segmented into easily separable characters/composite characters Statistical information about the height and width of each separated box is used to hypothesize whether a character box is composite In the second pass, the hypothesized composite characters are further segmented A recognition rate of 85 percent has been achieved on the segmented conjuncts The algorithm is designed to segment a pair of touching characters
TL;DR: An algorithm for a rotation invariant template matching method based on the combination of the projection method and Zernike moments is proposed and it is proposed that the matching candidates are selected using a computationally low cost feature.
Abstract: Template matching is the process of determining the presence and the location of a reference image or an object inside a scene image under analysis by a spatial cross-correlation process. Conventional cross-correlation type algorithms are computationally expensive. Furthermore, when the object in the image is rotated, the conventional algorithms cannot be used for practical purposes. In this paper, an algorithm for a rotation invariant template matching method based on the combination of the projection method and Zernike moments is proposed. The algorithm consists of two stages. In the first stage, the matching candidates are selected using a computationally low cost feature. Frequency domain calculation was adopted to reduce the computational cost for this stage. In the second stage, rotation invariant template matching is performed only on the matching candidates using Zernike moments.