Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374564
J. Girod, G. Martin, B. Heit, J. Brémont
The segmentation tool presented in this article takes advantage of orientation selection mechanisms which appear in the visual cortex, so that fine, well-situated edges are obtained in a grey-scale image. The search for the best spatial resolution limits our study to the central part of the fovea. The first part of this article deals with a schematic description of the path followed by visual information in the brain and, in particular, from the eye to the primary visual cortex. The model used accepts spatial grouping by the horizontal cells in Gaussian form, and takes advantage of the center-surround antagonism of the bipolar cells found on the retina. The model obtained, which is quite insensitive to noise, reconciles very well the different characteristics of the natural images without setting the parameters. The structure of operations employed in order to carry this out allows a real-time implementation on neural network or pipeline hardware to be envisaged.<>
{"title":"Image segmentation by the modelisation of the biological visual systems","authors":"J. Girod, G. Martin, B. Heit, J. Brémont","doi":"10.1109/ICNN.1994.374564","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374564","url":null,"abstract":"The segmentation tool presented in this article takes advantage of orientation selection mechanisms which appear in the visual cortex, so that fine, well-situated edges are obtained in a grey-scale image. The search for the best spatial resolution limits our study to the central part of the fovea. The first part of this article deals with a schematic description of the path followed by visual information in the brain and, in particular, from the eye to the primary visual cortex. The model used accepts spatial grouping by the horizontal cells in Gaussian form, and takes advantage of the center-surround antagonism of the bipolar cells found on the retina. The model obtained, which is quite insensitive to noise, reconciles very well the different characteristics of the natural images without setting the parameters. The structure of operations employed in order to carry this out allows a real-time implementation on neural network or pipeline hardware to be envisaged.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"46 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126133853","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374650
K. Koh, H. Beom, J.S. Kim, H. cho
For mobile robots to be autonomous, they should have essential functional capabilities such as determination of their current location and heading angle, path control in order to follow the desired path and local path planning for uncertain environments. This paper deals with the above three issues and illustrates how the artificial neural network can be utilized to solve such problems. This neural network-based navigation system offers a method of determining the mobile robot's position-a 3D landmark sensing system with neural estimator. It also offers a neural net-based feedforward controller designed to accurately track a desired path and a sensor-based local path planning capability to adapt to complex and changing environments. System software/hardware architecture to implement the above functional capabilities are discussed and some experimental and simulation results are illustrated to show the effectiveness of the proposed navigation system.<>
{"title":"A neural network-based navigation system for mobile robots","authors":"K. Koh, H. Beom, J.S. Kim, H. cho","doi":"10.1109/ICNN.1994.374650","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374650","url":null,"abstract":"For mobile robots to be autonomous, they should have essential functional capabilities such as determination of their current location and heading angle, path control in order to follow the desired path and local path planning for uncertain environments. This paper deals with the above three issues and illustrates how the artificial neural network can be utilized to solve such problems. This neural network-based navigation system offers a method of determining the mobile robot's position-a 3D landmark sensing system with neural estimator. It also offers a neural net-based feedforward controller designed to accurately track a desired path and a sensor-based local path planning capability to adapt to complex and changing environments. System software/hardware architecture to implement the above functional capabilities are discussed and some experimental and simulation results are illustrated to show the effectiveness of the proposed navigation system.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"154 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123774081","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374453
Bertram E. Shi, T. Roska, L. Chua
This paper introduces an analytic method to determine the sensitivity to random parameter variations of analog VLSI neural network architectures for linear image filtering. The authors compare the robustness of several different circuit architectures for low pass filtering. This method can also determine which components within a particular architecture should specified the most precisely.<>
{"title":"Random parameter variation in analog VLSI neural networks for linear image filtering","authors":"Bertram E. Shi, T. Roska, L. Chua","doi":"10.1109/ICNN.1994.374453","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374453","url":null,"abstract":"This paper introduces an analytic method to determine the sensitivity to random parameter variations of analog VLSI neural network architectures for linear image filtering. The authors compare the robustness of several different circuit architectures for low pass filtering. This method can also determine which components within a particular architecture should specified the most precisely.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"10 11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126754792","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374359
G. Heileman, M. Georgiopoulos, Juxin Hwang
A collection of results related to learning in ART1 networks is presented. These results are concerned primarily with the complexity of the learning process, rather than with the quality of the learned concepts. These results provide numerous insights into the operation of ART1 networks, and detail the conditions under which such networks can learn efficiently.<>
{"title":"A survey of learning results for ART1 networks","authors":"G. Heileman, M. Georgiopoulos, Juxin Hwang","doi":"10.1109/ICNN.1994.374359","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374359","url":null,"abstract":"A collection of results related to learning in ART1 networks is presented. These results are concerned primarily with the complexity of the learning process, rather than with the quality of the learned concepts. These results provide numerous insights into the operation of ART1 networks, and detail the conditions under which such networks can learn efficiently.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126834738","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374370
F. Allen, H. Caulfield
The PCNN developed by Johnson (1993) are syntactic pattern transformers. Hence their outputs are quite similar over a wide variety of "distortions". We show that we can convert a PCNN into an attractor system which, away from boundaries, produces point attractor icons which are ideal inputs to statistical pattern processors.<>
{"title":"Reentrant pulse coupled neural networks (PCNNs)","authors":"F. Allen, H. Caulfield","doi":"10.1109/ICNN.1994.374370","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374370","url":null,"abstract":"The PCNN developed by Johnson (1993) are syntactic pattern transformers. Hence their outputs are quite similar over a wide variety of \"distortions\". We show that we can convert a PCNN into an attractor system which, away from boundaries, produces point attractor icons which are ideal inputs to statistical pattern processors.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"28 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115053226","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374344
A. Sperduti, A. Starita
The labeling RAAM (LRAAM) is a neural network able to encode data structures in fixed size patterns, thus allowing the application of neural networks to structured domains. Moreover, the structures stored into an LRAAM can be accessed both by pointer and by content. In this paper we briefly discuss basic and generalized associative access procedures for the LRAAM. Basic procedures are obtained by transforming the LRAAM network into a BAM. Different constrained versions of the BAM are used depending on the key(s) used to retrieve information. Generalized procedures are implemented by generalized Hopfield networks (GHN) which are built both by composing the subset of weights compounding the LRAAM and according to the query used to retrieve information. Some examples for generalized procedures are given.<>
{"title":"On the access by content capabilities of the LRAAM","authors":"A. Sperduti, A. Starita","doi":"10.1109/ICNN.1994.374344","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374344","url":null,"abstract":"The labeling RAAM (LRAAM) is a neural network able to encode data structures in fixed size patterns, thus allowing the application of neural networks to structured domains. Moreover, the structures stored into an LRAAM can be accessed both by pointer and by content. In this paper we briefly discuss basic and generalized associative access procedures for the LRAAM. Basic procedures are obtained by transforming the LRAAM network into a BAM. Different constrained versions of the BAM are used depending on the key(s) used to retrieve information. Generalized procedures are implemented by generalized Hopfield networks (GHN) which are built both by composing the subset of weights compounding the LRAAM and according to the query used to retrieve information. Some examples for generalized procedures are given.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115247557","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374419
E. Tazaki, N. Inoue
In this paper, the authors first present a method for automated extraction of fuzzy rules using neural networks with a planar lattice architecture. The neural network is composed of three layers-input layer, hidden layer with a lattice architecture and output layer. In the hidden layer, the neurons are arranged in a lattice structure, with each neuron assigned a position in a lattice. Each neuron of the hidden layer is assigned a fuzzy proposition which composes a fuzzy rule. The network is learned structurally with generation/annihilation of neurons. After the rules learning process, one may extract simple fuzzy production rules from the network. Next, the authors extend the method to the cases of multi-dimensional rules. The authors apply the proposed method to generate the diagnostic rules for hernia of an intervertebral disc.<>
{"title":"A generation method for fuzzy rules using neural networks with planar lattice architecture","authors":"E. Tazaki, N. Inoue","doi":"10.1109/ICNN.1994.374419","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374419","url":null,"abstract":"In this paper, the authors first present a method for automated extraction of fuzzy rules using neural networks with a planar lattice architecture. The neural network is composed of three layers-input layer, hidden layer with a lattice architecture and output layer. In the hidden layer, the neurons are arranged in a lattice structure, with each neuron assigned a position in a lattice. Each neuron of the hidden layer is assigned a fuzzy proposition which composes a fuzzy rule. The network is learned structurally with generation/annihilation of neurons. After the rules learning process, one may extract simple fuzzy production rules from the network. Next, the authors extend the method to the cases of multi-dimensional rules. The authors apply the proposed method to generate the diagnostic rules for hernia of an intervertebral disc.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115579685","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374757
A. Bastian, J. Gasós
System identification can be divided into structure identification and parameter identification. In most system identification approaches the structure is presumed and only a parameter identification is performed to obtain the coefficients in the functional system. Unfortunately in many cases there is little knowledge about the system structure. The structure identification itself can be divided into two types: the identification of the input variables of the model and the input-output relation, here respectively named structure identification type I and type II. In this paper a black-box structure identification type I approach, using a feedforward neural network in combination with the regularity criterion in GMDH (group method of data handling) and a novel identification algorithm, is proposed.<>
{"title":"A type I structure identification approach using feedforward neural networks","authors":"A. Bastian, J. Gasós","doi":"10.1109/ICNN.1994.374757","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374757","url":null,"abstract":"System identification can be divided into structure identification and parameter identification. In most system identification approaches the structure is presumed and only a parameter identification is performed to obtain the coefficients in the functional system. Unfortunately in many cases there is little knowledge about the system structure. The structure identification itself can be divided into two types: the identification of the input variables of the model and the input-output relation, here respectively named structure identification type I and type II. In this paper a black-box structure identification type I approach, using a feedforward neural network in combination with the regularity criterion in GMDH (group method of data handling) and a novel identification algorithm, is proposed.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"64 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116124326","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374356
A. Krzyżak, L. Xu
Rather than studying the L/sub 2/ convergence rates of kernel regression estimators (KRE) and radial basis function (RBF) nets given in Xu-Krzyzak-Yuille (1992 & 1993), we study convergence properties of the mean integrated absolute error (MIAE) for KRE and RBF nets. It has been shown that MIAE of KRE and RBF nets can converge to zero as the size of networks and the size of the training sequence tend to /spl infin/, and that the upper bound for the convergence rate of MIAE is O(n-/sup /spl alpha/s/sub (2+s)/( /sub 2//spl alpha/+d)/) for approximating Lipschitz functions.<>
{"title":"Some results on L/sub 1/ convergence rate of RBF networks and kernel regression estimators","authors":"A. Krzyżak, L. Xu","doi":"10.1109/ICNN.1994.374356","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374356","url":null,"abstract":"Rather than studying the L/sub 2/ convergence rates of kernel regression estimators (KRE) and radial basis function (RBF) nets given in Xu-Krzyzak-Yuille (1992 & 1993), we study convergence properties of the mean integrated absolute error (MIAE) for KRE and RBF nets. It has been shown that MIAE of KRE and RBF nets can converge to zero as the size of networks and the size of the training sequence tend to /spl infin/, and that the upper bound for the convergence rate of MIAE is O(n-/sup /spl alpha/s/sub (2+s)/( /sub 2//spl alpha/+d)/) for approximating Lipschitz functions.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116381689","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374936
Y. Chiou, F. Lure, P. Ligomenides
A "Hybrid Lung Nodule Detection (HLND) system" based on artificial neural network architecture and interactive knowledge-base system is developed for object detection in noisy image environments. This paper describes the system architecture and its application to detection and classification of nodules in lung cancerous pulmonary radiology. The configuration of the HLND system includes the following processing phases: (1) pre-processing to enhance the figure-background contrast; (2) Morphology based quick selection of nodule object suspects based upon the most prominent feature of nodules; and (3) feature space determination and neural network based suspect fields reduction; (4) interactive knowledge base and knowledge fusion processing and final classification of nodule suspect fields. Preliminary results from the approach are also reported.<>
{"title":"Neural-knowledge base object detection in Hybrid Lung Nodule Detection (HLND) system","authors":"Y. Chiou, F. Lure, P. Ligomenides","doi":"10.1109/ICNN.1994.374936","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374936","url":null,"abstract":"A \"Hybrid Lung Nodule Detection (HLND) system\" based on artificial neural network architecture and interactive knowledge-base system is developed for object detection in noisy image environments. This paper describes the system architecture and its application to detection and classification of nodules in lung cancerous pulmonary radiology. The configuration of the HLND system includes the following processing phases: (1) pre-processing to enhance the figure-background contrast; (2) Morphology based quick selection of nodule object suspects based upon the most prominent feature of nodules; and (3) feature space determination and neural network based suspect fields reduction; (4) interactive knowledge base and knowledge fusion processing and final classification of nodule suspect fields. Preliminary results from the approach are also reported.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116409594","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：