NNSP-4.1

Experiments in Topic Indexing of Broadcast News using Neural Networks
Christoph Neukirchen, Daniel Willett, Gerhard Rigoll (Gerhard-Mercator-University, Duisburg)

The paper deals with the problem of extracting topic information from news show stories by statistical methods. It is shown that the traditional topic-dependent n-gram language modeling approach can be decomposed in order to apply neural networks for topic indexing. Two specific problems in training of these networks are addressed: a very sparse data distribution in the stories and the superposition of different topics in a story. The first problem is tackled by an integrated smoothing approach in the backpropagation method; an expansion of the neural network structure can be used to cope with topic mixtures in stories. Due to the efficient parameter sharing the application of neural networks results in a small improvement in topic indexing performance on a small corpus of broadcast news compared to the traditional topic-dependent n-gram method.

NNSP-4.2

Critical Input Data Channels Selection for Progressive Work Exercise Test by Neural network Sensitivity Analysis
Avni H Rambhia, Robb W Glenny, Jenq-Neng Hwang (University of Washington)

We aimed at training a neural network to classify stress test exercise data into one of three classes: normal, heart failure, or lung failure. Good classification accuracy was obtained using a backpropagation neural network architecture with one hidden layer during cross validation on data set of 110 vectors, when all 17 channels were used. We further aimed at determining which of these channels were critical to the decision making process. This was done through an input sensitivity analysis. Results showed that nine channels formed a critical superset of which possibly any eight could achieve almost perfect classification. We thus show that faster and more accurate classification may be obtained by input channel elimination due to dimension reduction of input space, which makes better generalization.

NNSP-4.3

Feature Selection Using General Regression Neural Networks for the Automatic Detection of Clustered Microcalcifications
Songyang Yu, Ling Guan (School of Electrical & Information Engineering, University of Sydney)

General regression neural networks (GRNNs) are proposed for selecting the most discriminating features for the automatic detection of clustered microcalcifications in digital mammograms. Previously, We have designed an image processing system for detecting clustered microcalcifications. The system uses wavelet coefficients and feed forward neural networks to identify possible microcalcification pixels and a set of structure features to locate individual microcalcifications. In this work, more features are extracted, and the most discriminating features are selected through the analysis of the GRNNs. The selected features are incorporated into our image processing system and applied to a database of 40 mammograms (Nijmegen database) containing 105 clusters of microcalcifications. Free response operating characteristics (FROC) curves are used to evaluate the performance. Results show that, by incorporating the proposed feature selection scheme, the performance of our system is improved significantly.

NNSP-4.4

Texture Edge Detection by Feature Encoding and Predictive Model
Jyh-Charn Liu, Gouchol Pok (Department of Computer Science, Texas A&M University)

Texture boundaries or edges are useful information for segmenting heterogeneous textures into several classes. Texture edge detection is different from the conventional edge detection that is based on the pixel-wise changes of gray level intensities, because textures are formed by patterned placement of texture elements over some regions. We propose a prediction-based texture edge detection method that includes encoding and prediction modules as its major components. The encoding module projects $n$-dimensional texture features onto a 1-dimensional feature map through SOFM algorithm to obtain scalar features, and the prediction module computes the predictive relationship of the scalar features with respect to their neighbors sampled from 8 directions. The variance of prediction errors is used as the measure for detection of edges. In the experiments with the micro-textures, our method has shown its effectiveness in detecting the texture edges.

NNSP-4.5

Edge Characterization Using a Model-Based Neural Network
Hau-san Wong (Department of Electrical Engineering, The University of Sydney, NSW 2006 Australia), Terry Caelli (Center for Mapping, The Ohio State University, Columbus, Ohio 43212, USA.), Ling Guan (Department of Electrical Engineering, The University of Sydney, NSW 2006, Australia)

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.

NNSP-4.6

LICENSE-PLATE LOCALIZATION BY USING VECTOR QUANTIZATION
Stefano Rovetta, Rodolfo Zunino (DIBE - Genoa University)

The paper describes a novel approach using Vector Quantization (VQ) to process vehicle images for automated identification. The VQ-based method yields superior quality in picture compression for archival purposes, and, at the same time, supports the localization of text regions in the image effectively. As opposed to standard approaches, VQ encoding gives some hint about the contents of image regions; such information is exploited to boost localization performance. The VQ system may be trained empirically from examples; this provides adaptiveness and on-field tuning facility. The approach has been tested in a real application and included satisfactorily into a complete system for vehicle identification.

NNSP-4.7

Neural Netwirk based Person Identification using EEG features.
Marios Poulos (University of Piraeus, Piraeus, Greece), maria Rangoussi (National Technical University of Athens, Athens, Greece), Nikolaos Alexandris (University of Piraeus, Piraeus, Greece)

A direct connection between the ElectroEncephaloGram (EEG) and the genetic information of an individual has been suspected and investigated by neurophysiologists and psychiatrists since 1960. However, most of this early, as well as more recent, research focuses on the classification of pathological EEG cases, aiming to construct tests for purposes of diagnosis. On the contrary, our work focuses on healthy individuals and aims to establish an one-to-one correspondence between the genetic information of the individual and certain features of the his/her EEG, as an intermediate step towards the further goal of developing a test for person identification based on features extracted from the EEG. Potential applications include, among others, information encoding and decoding and access to secure information. At the present stage, the proposed method uses spectral information extracted from the EEG non-parametrically, via the FFT, and employs a neural network (a Learning Vector Quantizer - LVQ) to classify unknown EEGs as belonging to one of a finite number of individuals. Correct classification scores ranging from 80% to 100% in experiments conducted on real data, show evidence that the EEG indeed carries genetic information and that the proposed method can be used to construct person identification tests based on the EEG.

NNSP-4.8

Sonar discrimination of cylinders from different angles using neural networks
Lars N Andersen (Department of Mathematical Modeling, Technical University of Denmark), Whitlow W Au (Marine Mammal Research Program, Hawaii Institute of Marine Biology, University of Hawaii), Jan Larsen, Lars K Hansen (Department of Mathematical Modeling, Technical University of Denmark)

This paper describes an underwater object discrimination system applied to recognize cylinders of various compositions from different angles. The system is based on a new combination of simulated dolphin clicks, simulated auditory filters and artificial neural networks. The model demonstrates its potential on real data collected from four different cylinders in an environment where the angles were controlled in order to evaluate the models capabilities to recognize cylinders independent of angles.

NNSP-4.9

A Neural Network based Transcoder for MPEG2 Video Compression
Hsin C Fu, Z. H Chen, Yeong Y Xu (Department of Computer Engineering, National Chiao Tung University), C. H Wang (Mentor Data Systems, Hsinchu, Taiwan)

In this paper, we proposed a neural network method for the high efficiency requatization in the design of a transcoder. In our design, there are two types of video bitrate control in the proposed transcoder. One is the global adjusting of quantizer scales in which the adjusting is based on the complication of the whole frame, the other is the adaptive adjusting of quantizer scales, that the adjusting is the complication of the current macroblock. From our experimental results, the prototype transcoder can achieve desirable bitrate (1.5 Mbps) with an acceptable image quality. In addition, we constructed a video multiplexer for PPV or NVOD applications on the proposed transcoder.

< NNSP-3

Last Update: February 4, 1999 Ingo Höntsch