Human target detection is known to be dependent on a number of components: one, basic electro-optics including image contrast, the target size, pixel resolution, and contrast sensitivity; two, target shape, image type and features, types of clutter; and three, context and task requirements. Here, we consider a Bayesian approach to investigating how these components contribute to target detection. To this end, we develop and compare three different formulations for contrast: mean contrast, perceptual contrast, and a Bayesian-based histogram contrast statistic. Results on past detection data show how the latter contrast measure correlates well with human performance factoring out all other dimensions. As for clutter, our findings show that with large targets, there are effectively no clutter effects. Furthermore, clutter does not have a major effect on detection when it is not contiguous with the target even when it is smaller. However, except for large targets, when the target is contiguous with the clutter, detection clearly decreases as a function of the similarity of target and clutter features-creating type of "clutter camouflage". This Bayesian formulation uses priors based on the contrast histogram statistics derived from all the images, the image context, and implies that human observers have adapted their criteria to fit with the image set, context, and task.

Graphs are a powerful and versatile data structure for pattern recognition. However, their flexibility brings inherent complexities for algorithms which seek to create or utilise graphs. In the context of computer vision, many feature detection algorithms can extract a suitable vertex set from an image. The creation of edges between these vertices presents new challenges, and effective methods for edge creation only exist for certain types of vertices such as points and regions. This paper presents a novel method for creating edges for image graphs, while supporting a wide array of vertex types. The presented method is principled, and its robustness is shown experimentally against a number of affine and projective transforms, as well as noise.

The autism spectrum disorder (ASD) is increasingly being recognized as a major public health issue which affects approximately 0.5-0.6% of the population. Promoting the general awareness of the disorder, increasing the engagement with the affected individuals and their carers, and understanding the success of penetration of the current clinical recommendations in the target communities, is crucial in driving research as well as policy. The aim of the present work is to investigate if Twitter, as a highly popular platform for information exchange, can be used as a data-mining source which could aid in the aforementioned challenges. Specifically, using a large data set of harvested tweets, we present a series of experiments which examine a range of linguistic and semantic aspects of messages posted by individuals interested in ASD. Our findings, the first of their nature in the published scientific literature, strongly motivate additional research on this topic and present a methodological basis for further work.

Low cost pervasive electrocardiogram (ECG) monitors is changing how sinus arrhythmia are diagnosed among patients with mild symptoms. With the large amount of data generated from long-term monitoring, come new data science and analytical challenges. Although traditional rule-based detection algorithms still work on relatively short clinical quality ECG, they are not optimal for pervasive signals collected from wearable devices—they don't adapt to individual difference and assume accurate identification of ECG fiducial points. To overcome these short-comings of the rule-based methods, this paper introduces an arrhythmia detection approach for low quality pervasive ECG signals. To achieve the robustness needed, two techniques were applied. First, a set of ECG features with minimal reliance on fiducial point identification were selected. Next, the features were normalized using robust statistics to factors out baseline individual differences and clinically irrelevant temporal drift that is common in pervasive ECG. The proposed method was evaluated using pervasive ECG signals we collected, in combination with clinician validated ECG signals from Physiobank. Empirical evaluation confirms accuracy improvements of the proposed approach over the traditional clinical rules.

Human actions have been widely studied for their potential application in various areas such as sports, pervasive patient monitoring, and rehabilitation. However, challenges still persist pertaining to determining the most useful ways to describe human actions at the sensor, then limb and complete action levels of representation and deriving important relations between these levels each involving their own atomic components. In this paper, we report on a motion encoder developed for the sensor level based on the need to distinguish between the shape of the sensor's trajectory and its temporal characteristics during execution. This distinction is critical as it provides a different encoding scheme than the usual velocity and acceleration measures which confound these two attributes of any motion. At the same time, we eliminate noise from sensors by comparing temporal and spatial indexing schemes and a number of optimal filtering models for robust encoding. Results demonstrate the benefits of spatial indexing and separating the shape and dynamics of a motion, as well as its ability to decompose complex motions into several atomic ones. Finally, we discuss how this specific type of sensor encoder bears on the derivation of limb and complete action descriptions.

Solutions for deriving the most consistent Bayesian gene regulatory network model from given data sets using evolutionary algorithms typically only result in locally optimal solutions. Further, due to genetic drift, merely increasing the size of the population does not overcome this limitation. In this paper, we propose a two-stage genetic algorithm that systematically searches the whole search space using frequent subgraph mining techniques. The approach finds representative patterns present in different local optimal solutions in the first stage and then combines these frequent subgraphs (motifs) in the second stage to converge to the global optima. We apply the algorithm to both synthetic and real life networks of yeast and E.coli and show the effectiveness of our approach.

Localising and aligning objects is a challenging task in computer vision that still remains largely unsolved. Utilising the syntactic power of graph representation, we define a relational string-graph matching algorithm that seeks to perform these tasks simultaneously. By matching the relations between vertices, where vertices represent high-level primitives, the relational string-graph is able to overcome the noisy and inconsistent nature of the vertices themselves. For each possible relation correspondence between two graphs, we calculate the rotation, translation, and scale parameters required to transform a relation into its counterpart. We plot these parameters in 4D space and use Gaussian mixture models and the expectation-maximisation algorithm to estimate the underlying parameters. Our method is tested on face alignment and recognition, but is equally (if not more) applicable for generic object alignment.

When tracking a target using phase-only signal returns, range ambiguities are a major issue. In this work, a look-up table between phase measurement space and target location space is constructed for phase measurement mapping. Solving such problems via solutions to Diophantine equations has been used to locate candidate locations. Here we show how such target location ambiguity can also be resolved over time when the underlying target is in motion and where issues of clutter are treated via a phase distribution discrimination method. That is, a probability density function of the ambiguous phase-only measurement that takes both sensor noise and target motion distributions into account is derived based on directional statistics. This approach to solving phase ambiguity under significant clutter conditions has promising results.

The authors examine the problem of segmenting foreground objects in live video when background scene textures change over time. In particular, we formulate background subtraction as minimizing a penalized instantaneous risk functional--yielding a local online discriminative algorithm that can quickly adapt to temporal changes. We analyze the algorithm's convergence, discuss its robustness to nonstationarity, and provide an efficient nonlinear extension via sparse kernels. To accommodate interactions among neighboring pixels, a global algorithm is then derived that explicitly distinguishes objects versus background using maximum a posteriori inference in a Markov random field (implemented via graph-cuts). By exploiting the parallel nature of the proposed algorithms, we develop an implementation that can run efficiently on the highly parallel graphics processing unit (GPU). Empirical studies on a wide variety of datasets demonstrate that the proposed approach achieves quality that is comparable to state-of-the-art offline methods, while still being suitable for real-time video analysis ( ≥ 75 fps on a mid-range GPU).

3D insect models are useful to overcome viewing angle variations and self-occlusions in computer-assisted insect taxonomy for electronic field guides. The acquisition of 3D information is, however, unreliable due to the flexibility and small size of the insect bodies. This paper explores how to infer 3D insect models from a single 2D insect image, which will assist both insect description and identification. The 3D structure of the insect body is modeled from two geometric primitives, generalized cylinders and deformable ellipsoids. The primitives are fitted and warped based on both edge and medial axis constraints of the 2D image. Individualized 3D models are then built to approximate the insect structure. The proposed approach results in seemingly useful 3D insect models capable of representing the major morphological characteristics for a variety of insects with different body types. This method could be a helpful assistance for computer-assisted insect taxonomy and insect identification by entomologists and the public.

Micropattern based image representation and recognition, e.g. Local Binary Pattern (LBP), has been proved successful over the past few years due to its advantages of illumination tolerance and computational efficiency. However, LBP only encodes the first-order radial-directional derivatives of spatial images and is inadequate to completely describe the discriminative features for classification. This paper proposes a new Circular Derivative Pattern (CDP) which extracts high-order derivative information of images along circular directions. We argue that the high-order circular derivatives contain more detailed and more discriminative information than the first-order LBP in terms of recognition accuracy. Experimental evaluation through face recognition on the FERET database and insect classification on the NICTA Biosecurity Dataset demonstrated the effectiveness of the proposed method.

In this paper, we aim at presenting a survey on automatic absorption recovery methods for imaging spectroscopy. We commence by viewing the algorithms in the literature from a technical perspective and presenting an overview of the derivative analysis, fingerprint, and maximum modulus wavelet transform techniques. In addition to these methods, we also present a novel absorption recovery approach based upon unimodal regression and continuum removal. With this technical review of the methods under study, we perform a complexity analysis and examine the implementation issues pertaining to each of the alternatives. We show how detected absorption bands can be used for purposes of material identification. We conclude this paper by providing a performance study and providing identification results on hyperspectral imagery. To this end, we make use of a number of distance measures to evaluate the quality of the recovered absorptions, as compared to continuum-removed spectra.

We present a simple, principled approach to detecting foreground objects in video sequences in real-time. Our method is based on an on-line discriminative learning technique that is able to cope with illumination changes due to discontinuous switching, or illumination drifts caused by slower processes such as varying time of the day. Starting from a discriminative learning principle, we derive a training algorithm that, for each pixel, computes a weighted linear combination of selected past observations with time-decay. We present experimental results that show the proposed approach outperforms existing methods on both synthetic sequencse and real video data.

In this paper we present a simple but novel technique to recover 3D temperature map from a pair of calibrated thermal cameras. The method is also useful in estimating depth of an object in the dark environment

We present two new algorithms for online learning in reproducing kernel Hilbert spaces. Our first algorithm, ILK (implicit online learning with kernels), employs a new, implicit update technique that can be applied to a wide variety of convex loss functions. We then introduce a bounded memory version, SILK (sparse ILK), that maintains a compact representation of the predictor without compromising solution quality, even in non-stationary environments. We prove loss bounds and analyze the convergence rate of both. Experimental evidence shows that our proposed algorithms outperform current methods on synthetic and real data.

This paper describes a novel solution to the rigid point pattern matching problem in Euclidean spaces of any dimension. Although we assume rigid motion, jitter is allowed. We present a noniterative, polynomial time algorithm that is guaranteed to find an optimal solution for the noiseless case. First, we model point pattern matching as a weighted graph matching problem, where weights correspond to Euclidean distances between nodes. We then formulate graph matching as a problem of finding a maximum probability configuration in a graphical model. By using graph rigidity arguments, we prove that a sparse graphical model yields equivalent results to the fully connected model in the noiseless case. This allows us to obtain an algorithm that runs in polynomial time and is provably optimal for exact matching between noiseless point sets. For inexact matching, we can still apply the same algorithm to find approximately optimal solutions. Experimental results obtained by our approach show improvements in accuracy over current methods, particularly when matching patterns of different sizes.

In this paper, the optimizations of three fundamental components of image understanding: segmentation/annotation, 3D sensing (stereo) and 3D fitting, are posed and integrated within a Bayesian framework. This approach benefits from recent advances in statistical learning which have resulted in greatly improved flexibility and robustness. The first two components produce annotation (region labeling) and depth maps for the input images, while the third module integrates and resolves the inconsistencies between region labels and depth maps to fit most likely 3D models. To illustrate the application of these ideas, we have focused on the difficult problem of fitting individual tree models to tree stands which is a major challenge for vision-based forestry inventory systems.

In this study, we investigated the use of Bayesian networks for inferring tropical dry forest leaf area index (LAI) from satellite imagery in dry and wet seasons. LAI was chosen as the variable of interest because leaf area is the exchange surface between the photosynthetically active component of the canopy and the atmosphere. Initial network estimates were obtained from ground truth plot data with known forest structure, LAI, and satellite reflectance in the red and near-infrared bands (as observed by the Landsat 7 Enhanced Thematic Mapper Plus sensor). We tested the performance of the Bayesian networks with scoring rules and also with confidence and surprise scores. We evaluated the networks on a per-pixel basis and created both LAI maps of the study area as well predicted the probability maps for the highest LAI states. Results not only demonstrate the predictive power of a Bayesian network but also its explanatory power which is far beyond what is typically available with current pixel classifier approaches such as spectral vegetation indices or other approaches such as neural networks.

Graph eigenspaces have been used to encode many different properties of graphs. In this paper we explore how such methods can be used for solving inexact graph matching (the matching of sets of vertices in one graph to those in another) having the same or different numbers of vertices. In this case we explore eigen-subspace projections and vertex clustering (EPS) methods. The correspondence algorithm enables the EPC method to discover a range of correspondence relationships from one-to-one vertex matching to that of inexact (many-to-many) matching of structurally similar subgraphs based on the similarities of their vertex connectivities defined by their positions in the common subspace. Examples in shape recognition and random graphs are used to illustrate this method.

In this paper, we show how inexact graph matching (that is, the correspondence between sets of vertices of pairs of graphs) can be solved using the renormalization of projections of the vertices (as defined in this case by their connectivities) into the joint eigenspace of a pair of graphs and a form of relational clustering. An important feature of this eigenspace renormalization projection clustering (EPC) method is its ability to match graphs with different number of vertices. Shock graph-based shape matching is used to illustrate the model and a more objective method for evaluating the approach using random graphs is explored with encouraging results.

We examine a new double-layered Markov random field probabilistic framework for stereo matching and use belief propagation for approximate inference. Our initial experimental results are promising and future developments are discussed.

A framework for diffusion of color images is presented. The method is based on the theory of thermodynamics of irreversible transformations which provides a suitable basis for designing correlations between the different color channels. More precisely, we derive an equation for color evolution which comprises a purely spatial diffusive term and a nonlinear term that depends on the interactions among color channels over space. We apply the proposed equation to images represented in several color spaces, such as RGB, CIELAB, Opponent colors, and IHS.

International Journal of Pattern Recognition and Artificial IntelligenceVol. 15, No. 01, pp. 1-3 (2001) SPECIAL ISSUE: Hidden Markov Models in Vision; Edited by H. Bunke and T. CaelliNo AccessEDITORIALHORST BUNKE and TERRY CAELLIHORST BUNKEInstitut für Informatik und Angewandte Mathematik, Univerität Bern, Neubrückstrasse 10, CH-3012 Bern, Switzerland and TERRY CAELLIDepartment of Psychology, University of Alberta, P220, Biological Sciences Building, Edmonton, Alberta, Canada T6G 2E9, Canadahttps://doi.org/10.1142/S0218001401000721Cited by:2 Next AboutSectionsPDF/EPUB ToolsAdd to favoritesDownload CitationsTrack CitationsRecommend to Library ShareShare onFacebookTwitterLinked InRedditEmail FiguresReferencesRelatedDetailsCited By 2Beyond Traditional Kernels: Classification in Two Dissimilarity-Based Representation SpacesElzbieta Pekalska and Robert P. W. Duin1 Nov 2008 | IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), Vol. 38, No. 6Diagnostic tools for evaluating and updating hidden Markov modelsBrendan McCane and Terry Caelli1 Jul 2004 | Pattern Recognition, Vol. 37, No. 7 Recommended Vol. 15, No. 01 Metrics History PDF download

This paper is concerned with an application of Hidden Markov Models (HMMs) to the generation of shape boundaries from image features. In the proposed model, shape classes are defined by sequences of "shape states" each of which has a probability distribution of expected image feature types (feature "symbols").The tracking procedure uses a generalization of the well-known Viterbi method by replacing its search by a type of "beam-search" so allowing the procedure, at any time, to consider less likely features (symbols) as well the search for an instantiable optimal state sequences. We have evaluated the model's performance on a variety of image and shape types and have also developed a new performance measure defined by an expected Hamming distance between predicted and observed symbol sequences. Results point to the use of this type of model for the depiction of shape boundaries when it is necessary to have accurate boundary annotations as, for example, occurs in Cartography.

The evolution of information in images undergoing fine-to-coarse anisotropic transformations is analyzed by using an approach based on the theory of irreversible transformations. In particular, we show that, when an anisotropic diffusion model is used, local variation of entropy production over space and scale provides the basis for a general method to extract relevant image features.

Loss of information in images undergoing fine-to-coarse image transformations is analyzed by using an approach based on the theory of irreversible transformations. The method is applied to isotropic and anisotropic diffusion and an algorithm based on entropy variation is presented that extracts relevant features of the image and provides a method to discriminate between smooth, textured and edge-type regions.

We generalize to colour images a method to calculate the local information content in an image. The method is based on the theory of thermodynamics of irreversible transformations and relies on fine-to-coarse transformations, such as diffusion processes. To this end, we discuss an extension to multi-valued images of isotropic diffusion, where cross-interactions between the different colour channels are allowed. Results achieved in simulation are presented and discussed.

In this paper we consider how basic image feature extraction can be posed in terms of the development of a class of machine learning algorithms which are capable of tracking and predicting how humans perform tasks such as contour extraction and shape boundary tracking. In particular we consider how both recursive modular neural networks (RMNN) and hidden Markov models (HMM) can provide reasonably robust models for such tasks. Finally, we investigate how well they can predict human performance and so provide a reasonable basis for the development of more efficient and reliable human-machine annotation systems. Examples in sketching and cartography are discussed.

In this paper we discuss uses of image segmentation, feature extraction and Bayesian networks for identifying buildings in digital orthophotos and the utilisation of the results for the automated computation of building statistics. Our work differs from previous attempts in a number of ways. Firstly, image segmentation is accomplished using an adaptive multi-scale method which brings together region and edge information to segment the image into regions. Secondly, automated building feature extraction (e.g. corners) is optimised to fit with expert human annotation performance. The third aspect of this work is the exploration of Bayesian networks as a method for fusing available information (ranging from corner information to solar angles as indicators of shadow location) to classify segmented regions as corresponding to buildings or not. Such processes then permit the automatic compilation of building statistics.

This paper further explores the uses of Bayesian Networks for detecting buildings from digital orthophotos. This work differs from current research in building detection in so far as it utilizes the ability of Bayesian Networks to provide probabilistic methods for evidence combination and, via training, to determine how such evidence should be weighted to maximize classification. In this vein, then, we have also utilized expert performance to not only configure the network values but also to adapt the feature extraction pre-processing units to fit human behavior as closely as possible. Results from digital orthophotos of the Washington DC area prove that such an approach is feasible, robust and worth further analysis.

This paper describes a low-cost interactive active monocular range finder and illustrates the effect of introducing interactivity to the range acquisition process. The range finder consists of only one camera and a laser pointer to which three LEDs are attached. When a user scans the laser along surfaces of objects, the camera captures the image of spots (one from the laser and the others from LEDs), and triangulation is carried our using the camera's viewing direction and the optical axis of the laser. The user interaction allows the range finder to acquire range data in which the sampling rate varies across the object depending on the underlying surface structures. Moreover the processes of separating objects from the background and/or finding parts in the object can be achieved using the operator's knowledge of the objects.

In this paper, we investigate the feasibility of characterizing significant image edges using a model-based neural network with modular architecture. Instead of employing traditional mathematical models for characterization, we ask human users to select what they regard as significant features on an image, and then incorporate these selected edges directly as training examples for the network. Unlike conventional edge detection schemes where decision thresholds have to be specified, the current NN-based edge characterization scheme implicitly represents these decision parameters in the form of network weights which are updated during the training process. Experiments have confirmed that the resulting network is capable of generalizing this previously acquired knowledge to identify important edges in images not included in the training set. Most importantly, the current approach is very robust against noise contaminations, such that no re-training of the network is required when it is applied to noisy images.

This paper describes object-centered symbolic representation and distributed matching strategies of 3D objects in a schematic form which occur in engineering drawings and maps. The object-centered representation has a hierarchical structure and is constructed from symbolic representations of schematics. With this representation, two independent schematics representing the same object can be matched. We also consider matching strategies using distributed algorithms. The object recognition is carried out with two matching methods: (1) matching between an object model and observed data at the lowest level of the hierarchy, and (2) constraints propagation. The first is carried out with symbolic Hopfield-type neural networks and the second is achieved via hierarchical winner-takes-all algorithms.

Employing a thermodynamical model applied to scale-space transformations, we derive a method to encode different types of image features, as regions of different information content. Information is locally defined in terms of activity of the density of entropy production along scales. An activity map is formed, which represents the conspicuity or saliency of each point of the image. According to the different levels of activity, labelled as high, medium or low, we can extract basic parts of the image by collecting points of similar saliency.

This paper deals with a novel way for representing and computing image features encapsulated within different regions of scale-space. Employing a thermodynamical model for scale-space generation, the method derives features as those corresponding to "entropy rich" image regions where, within a given range of spatial scales, the entropy gradient remains constant. Different types of image features, defining regions of different information content, are accordingly encoded by such regions within different bands of spatial scale.

This paper considers neural computing models for information processing in terms of collections of subnetwork modules. Two approaches to generating such networks are studied. The first approach includes networks with functionally independent subnetworks, where each subnetwork is designed to have specific functions, communication, and adaptation characteristics. The second approach is based on algorithms that can actually generate network and subnetwork topologies, connections, and weights to satisfy specific constraints. Associated algorithms to attain these goals include evolutionary computation and self-organizing maps. We argue that this modular approach to neural computing is more in line with the neurophysiology of the vertebrate cerebral cortex, particularly with respect to sensation and perception. We also argue that this approach has the potential to aid in solutions to large-scale network computational problems - an identified weakness of simply defined artificial neural networks.

We describe a color image transform which is capable of encoding spatio-chromatic image features. The method relies on the development of physical interpretations of complex image Fourier transforms allowing the direct extraction of spatio-chromatic structures from images. The process is analysed using an elliptical representation in order to determine filter characteristics and responses. Both a spatial and chromatic filter are developed and used on test images with results showing the ease with which this technique can be utilised.

We explore the use of natural language understanding and image processing to index and query American Football tapes. We present a model for representing spatio-temporal characteristics of multiple objects in dynamic scenes in this domain, and a recognition system which uses the model to recognise American Football plays.

This paper explores an efficient technique for the extraction of common subtrees in decision trees. The method is based on a Suffix Tree string matching process and the algorithm is applied to the problem of finding common decision rules in path planning.

The authors present an adaptive multi-scale algorithm using edge and region information for segmenting intensity images into closed regions. The need for segmentation is determined by region statistics and segmentation is actually performed using edge based information. Results are shown for a number of images displaying significant improvements over mono-scale segmentation.

A new learning algorithm, the Consolidated Learning Algorithm based on Relational Evidence Theory (CLARET) is presented, which relies on inductive logic programming and graph matching techniques. In relational evidence theory, two different approaches to evidential learning are consolidated in how they apply to generalising within relational data structures. Attribute-based discrimination (decision trees) is integrated with part-based interpretation (graph matching) for evaluating and updating representations in spatial domains. This allows an interpretation stage to be incorporated into the generalization process. A systematic approach to finding interpretations is demonstrated for an online, hand drawn, schematic diagram and symbols recognition system. The approach uses an adaptive representational bias and search strategy during learning by efficiently grounding the learning procedures in the relational spatial constraints of their application.

Machine learning has been applied to many problems related to scene interpretation. It has become clear from these studies that it is important to develop or choose learning procedures appropriate for the types of data models involved in a given problem formulation. In this paper, we focus on this issue of learning with respect to different data structures and consider, in particular, problems related to the learning of relational structures in visual data. Finally, we discuss problems related to rule evaluation in multi-object complex scenes and introduce some new techniques to solve them.

In this paper we further explore the use of machine learning (ML) for the recognition of 3D objects in isolation or embedded in scenes. Of particular interest is the use of a recent ML technique (specifically CRG — Conditional Rule Generation) which generates descriptions of objects in terms of object parts and part-relational attribute bounds. We show how this technique can be combined with intensity-based model and scene–views to locate objects and their pose. The major contributions of this paper are: the extension of the CRG classifier to incorporate fuzzy decisions (FCRG), the application of the FCRG classifier to the problem of learning 3D objects from 2D intensity images, the study of the usefulness of sparse depth data in regards to recognition performance, and the implementation of a complete object recognition system that does not rely on perfect or synthetic data. We report a recognition rate of 80% for unseen single object scenes in a database of 18 non-trivial objects.

We consider how machine learning can be used to help solve the problem of identifying objects or structures composed of parts in complex scenes. We first discuss a conditional rule generation technique that is designed to describe structures using part attributes and their relations. We then show how the resultant rules can be used for region labeling and examine constraint propagation techniques for improving rule-based object classification.

We discuss automatic rule generation techniques for learning relational properties of 2D visual patterns and 3D objects from training samples where the observed feature values are continuous. In particular, we explore a conditional rule generation method that defines patterns (or objects) in terms of ordered lists of bounds on unary (pattern part) and binary (part relation) features. The technique, termed conditional rule generation, was developed to integrate relational structure representations of patterns and the generalization characteristics of evidenced-based systems. We show how this technique can be used for recognition of complex patterns and of objects in scenes. Further, we show the extent to which the learned rules can identify patterns and objects that have undergone nonrigid distortions.

In this paper, we present a new off-line word recognition system that is able to recognize unconstrained handwritten words using grey-scale images. This is based on structural and relational information in the handwritten word. We use Gabor filters to extract features from the words, and then use an evidence-based approach for word classification. A solution to the Gabor filter parameter estimation problem is given, enabling the Gabor filter to be automatically tuned to the word image properties. We also developed two new methods for correcting the slope of the handwritten words. Our experiments show that the proposed method achieves good recognition rates compared to standard classification methods.

This paper describes robust hierarchical modeling of the image amplitude spectrum via sets of bivariate Gaussian functions which involves: adaptive determination of a low-pass filter, clustering of residual high-pass spectrum, and parametric encoding of separate spectral segments. Based on this modeling a small set of Gabor filters tuned to the channel of high activity in the image Fourier spectrum is determined and used to generate feature images for texture segmentation. In the segmentation algorithm a similar robust modeling procedure is applied to encode histograms of the feature images as mixtures of univariate Gaussians.

In this paper, Kohonen's self-organizing feature map is modified by a novel technique of allowing the neurons in the feature map to compete in a selective manner. The selective competition is achieved by grating the N-dimensional feature space using a spatial frequency and setting a criterion for the neurons to compete based on the region in which the input pattern resides. The spatial grating and selective competition are achieved by introducing a gated neuronal architecture in the feature map. As the selection criterion changes with time, it generates a time sequence of winning node indexes providing more input information and potentially allowing higher classification performance. These time sequences are then used to predict the class label of the input pattern more accurately. Three possible class label prediction algorithms are formulated based on evidential reasoning method and Bayes conditional probability theorem. These are tested on real world 8-class texture and a synthetic 12-class 3D object recognition problems. The classification performance is then compared with the results obtained by using a standard statistical linear discriminant analysis.

In this paper it is shown that the spatio-temporal signature generated for any input pattern on a topologically ordered feature map using a gated neuronal architecture is invariant over a neighbourhood of the input pattern provided the input patterns lie in the interior of the decision space and the regions of competition created by n-dimensional spatial grating function at any given spatial frequency are open. The spatio-temporal signature in a Gated Neuronal Architecture uniquely represents a collection of disjoint regions in the feature space. For pattern classification the labeling of the set of disjoint regions represented by the spatio-temporal signature is obtained by using Bayes conditional probabilities. Simulation results indicate improved performance.>

In this correspondence, the application of covariance techniques to surface representation of 3-D objects is discussed and such ways of computing surface geometry are compared with traditional methods using differential geometry. It is shown how the covariance method provides surface descriptors that are invariant to rigid motions without explicitly using surface parameterizations or derivatives. Analogous covariance operators for both the Gauss and Weingarten maps are defined and a range image segmentation technique is presented that labels pixels as jump or crease discontinuities or planar, parabolic or curved region types.>

The authors have extended the evidence-based object recognition system of Jain and Hoffman (1988) to include some new view-independent features, a new optimized rule generation procedure based upon minimum entropy clustering and a neural network which estimates optimal evidence weights and provides an associated matching procedure. This approach provides an objective definition of the difficulty of an object recognition problem. The authors also evaluate the procedures and performance of the system with two sets of CAD (range) models.>

The authors describe a hybrid system which uses multiple depth resources, rule-based depth interpolation and predictive rendering to generate accurate and dense depth maps. Initial depth estimates are obtained from multiple views (shape-from-generalised-stereo) and by shape-from-focus. The resultant sparse depth maps are then interpolated symbolically. The interpolated depth map for each region is corrected by inverse rendering. It is shown that nontrivial integration of depth resources reduces many of the problems normally associated with considering each depth resource individually.>

Extensions to current models for texture segmentation are presented, in which the underlying detector (filter) mechanisms are allowed to adapt to the incoming signal in terms of their dynamical response range and associativities. This system converges on new texton (B. Julesz, 1981) profiles of minimal dimensionality that are used to classify texture regions by a minimum distance classifier in the texture feature space. The three processes of convolution, cooperativity, and classification are individually analyzed and compared with some observations from human texture discrimination experiments.>

Questions concerning the number, tuning widths, and phase sensitivity of two-dimensional narrow-band energy processing units, particularly in central vision, are reviewed. Results with grating and two-dimensional images are discussed, particularly within the context of energy processing and information reduction strategies for the human observer. Finally, a consequent principle of invariance is defined and shown to hold with such filter units.