From the Publisher:
This is the first comprehensive treatment of feed-forward neural networks from the perspective of statistical pattern recognition. After introducing the basic concepts, the book examines techniques for modelling probability density functions and the properties and merits of the multi-layer perceptron and radial basis function network models. Also covered are various forms of error functions, principal algorithms for error function minimalization, learning and generalization in neural networks, and Bayesian techniques and their applications. Designed as a text, with over 100 exercises, this fully up-to-date work will benefit anyone involved in the fields of neural computation and pattern recognition.

/pdf/neural-networks-for-pattern-recognition-34puigiau1.pdf

Neural networks for pattern recognition

Methods for digital-image texture analysis are reviewed based on available literature and research work either carried out or supervised by the authors The review has been prepared on request of Dr Richard Lerski, Chairman of the Management Committee of the COST B11 action "Quantitation of Magnetic Resonance Image Texture"

/pdf/texture-analysis-methods-a-review-3pph5x6k81.pdf

Texture Analysis Methods - A Review

An efficient, federated Kalman filter is developed for use in distributed multisensor systems. The design accommodates sensor-dedicated local filters, some of which use data from a common reference subsystem. The local filters run in parallel, and provide sensor data compression via prefiltering. The master filter runs at a selectable reduced rate, fusing local filter outputs via efficient square root algorithms. Common local process noise correlations are handled by use of a conservative matrix upper bound. The federated filter yields estimates that are globally optimal or conservatively suboptimal, depending upon the master filter processing rate. This design achieves a major improvement in throughput (speed), is well suited to real-time system implementation, and enhances fault detection, isolation, and recovery capability. >

Federated square root filter for decentralized parallel processors

This note deals with the effect of the common process noise on the fusion (combination) of the state estimates of a target based on measurements obtained by two different sensors. This problem arises in a multisensor environment where each sensor has its information processing (tracking) subsystem. In the case of an ?-s tracking filter the effect of the process noise is that, over a wide range of its variance, the uncertainty area corresponding to the fused estimates is about 70 percent of the single-sensor uncertainty area as opposed to 50 percent obtained if the dependence is ignored.

The Effect of the Common Process Noise on the Two-Sensor Fused-Track Covariance

We develop simple but effective fuzzy-rule based models of complex systems from input-output data. We introduce a simple fuzzy-neural network for modeling systems, and we prove that it can represent any continuous function over a compact set. We introduce "fuzzy curves" and use them to: 1) identify significant input variables, 2) determine model structure, and 3) set the initial weights in the fuzzy-neural network model. Our method for input identification is computationally simple and, since we determine the proper network structure and initial weights in advance, we can train the network rapidly. Viewing the network as a fuzzy model gives insight into the real system, and it provides a method to simplify the neural network. >

A new approach to fuzzy-neural system modeling

In this paper we consider the problem of combining and updating estimates that may have been generated in a distributed fashion or may represent estimates, generated at different times, of the same process sample path. The first of these cases has applications in decentralized estimation, while the second has applications in updating maps of spatially-distributed random quantities given measurements along several tracks. The method of solution for the second problem uses the result of the first, and the similarity in the formulation and solution of these problems emphasizes the conceptual similarity between many problems in decentralized control and in the analysis of random fields.

/pdf/combining-and-updating-of-local-estimates-and-regional-maps-22baebl2hq.pdf

Combining and updating of local estimates and regional maps along sets of one-dimensional tracks

The standard backpropagation-based multilayer perceptron training algorithm suffers from a slow asymptotic convergence rate. Sophisticated nonlinear least-squares and quasi-Newton optimization techniques are used to construct enhanced multilayer perceptron training algorithms, which are then compared to the backpropagation algorithm in the context of several example problems. In addition, an integrated approach to training and architecture selection that uses the described enhanced algorithms is presented, and its effectiveness illustrated in the context of synthetic and actual pattern recognition problems. >

Enhanced training algorithms, and integrated training/architecture selection for multilayer perceptron networks

The author formulates a novel segmentation algorithm which combines the use of Markov random field models for image-modeling with the use of the discrete wavepacket transform for image analysis. Image segmentations are derived and refined at a sequence of resolution levels, using as data selected wave-packet transform images or "channels". The segmentation algorithm is compared with nonmultiresolution Markov random field-based image segmentation algorithms in the context of synthetic image example problems, and found to be both significantly more efficient and effective. >

A combined Markov random field and wave-packet transform-based approach for image segmentation

Martingale decomposition techniques are used to derive Markovian models for the error in smoothed estimates of processes described by linear models driven by white noise. These models, together with some simple Hilbert space decomposition ideas, provide a simple unified framework for examining a variety of problems involving the efficient assimilation of spatial data, which we refer to as mapping problems. Algorithms for several different mapping problems are derived. A specific example of map updating for a two-dimensional random field is included.

Smoothing error dynamics and their use in the solution of smoothing and mapping problems

A novel anomaly screening algorithm is described which makes use of a regression diagnostic associated with the fitting of Markov Random Field (MRF) models. This regression diagnostic quantifies the extent to which a given neighborhood of pixels is atypical, relative to local background characteristics. The screening algorithm consists first in the calculation of an MRF-based anomoly statistic values. Next, 'blob' features, such as pixel count and maximal pixel intensity are calculated, and ranked over the image, in order to 'filter' the blobs to some final subset of most likely candidates. Receiver operating characteristics obtained from applying the above described screening algorithm to the detection of minelike targets in high- and low-frequency side-scan sonar imagery are presented together with results obtained from other screening algorithms for comparison, demonstrating performance comparable to trained human operators. In addition, real-time implementation considerations associated with each algorithmic component of the described procedure are identified.

Martin G. Bello

Papers

Combining and updating of local estimates and regional maps along sets of one-dimensional tracks

Enhanced training algorithms, and integrated training/architecture selection for multilayer perceptron networks

A combined Markov random field and wave-packet transform-based approach for image segmentation

Smoothing error dynamics and their use in the solution of smoothing and mapping problems

Markov random-field-based anomaly screening algorithm