Pub Date : 2002-07-22DOI: 10.1109/CNNA.2002.1035107
J. Poikonen, A. Paasio
Mathematical morphology provides tools for many image processing tasks. In this paper we discuss the implementation of grayscale morphological operators of erosion, dilation and reconstruction with a hardware efficient ranked order filter circuit. By using dedicated hardware for these basic operations a higher performance of processing more complex functions in a massively parallel processor array can be achieved. Because the circuit realization of the ranked order filter used is very compact, the area required for one processing cell can be kept low. Simulations of the operation were performed with a 0.18 /spl mu/m digital CMOS technology.
{"title":"Implementing grayscale morphological operators with a compact ranked order extractor circuit","authors":"J. Poikonen, A. Paasio","doi":"10.1109/CNNA.2002.1035107","DOIUrl":"https://doi.org/10.1109/CNNA.2002.1035107","url":null,"abstract":"Mathematical morphology provides tools for many image processing tasks. In this paper we discuss the implementation of grayscale morphological operators of erosion, dilation and reconstruction with a hardware efficient ranked order filter circuit. By using dedicated hardware for these basic operations a higher performance of processing more complex functions in a massively parallel processor array can be achieved. Because the circuit realization of the ranked order filter used is very compact, the area required for one processing cell can be kept low. Simulations of the operation were performed with a 0.18 /spl mu/m digital CMOS technology.","PeriodicalId":387716,"journal":{"name":"Proceedings of the 2002 7th IEEE International Workshop on Cellular Neural Networks and Their Applications","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131163437","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 : 2002-07-22DOI: 10.1109/CNNA.2002.1035093
T. Nakaguchi, K. Omiya, M. Tanaka
Hysteresis cellular neural networks are one of artificial neural networks which work effectively against large scale problems. In the previous work, remarkable methods have never been developed to overcome the defects of hysteresis cellular neural networks. We then propose a novel architecture for combinatorial optimization problems to overcome them. Experimental results indicate the efficiency of the architecture.
{"title":"Hysteresis cellular neural networks for solving combinatorial optimization problems","authors":"T. Nakaguchi, K. Omiya, M. Tanaka","doi":"10.1109/CNNA.2002.1035093","DOIUrl":"https://doi.org/10.1109/CNNA.2002.1035093","url":null,"abstract":"Hysteresis cellular neural networks are one of artificial neural networks which work effectively against large scale problems. In the previous work, remarkable methods have never been developed to overcome the defects of hysteresis cellular neural networks. We then propose a novel architecture for combinatorial optimization problems to overcome them. Experimental results indicate the efficiency of the architecture.","PeriodicalId":387716,"journal":{"name":"Proceedings of the 2002 7th IEEE International Workshop on Cellular Neural Networks and Their Applications","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123496705","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 : 2002-07-22DOI: 10.1109/CNNA.2002.1035084
L. Kék
This paper proposes an improved method for systematic decomposition of Boolean operators into a sequence of simpler ones. The improvement has two main components: (i) a sufficient condition for decreasing the number of possible child-templates during decomposition; (ii) pointing out the template element, the elimination of which results in the possibly maximum increment of the robustness value of the template. Examples are presented to demonstrate the effectiveness of the proposed method, whose advantages and limitations are also discussed.
{"title":"Improvement of the method for uncoupled binary input-output CNN template decomposition","authors":"L. Kék","doi":"10.1109/CNNA.2002.1035084","DOIUrl":"https://doi.org/10.1109/CNNA.2002.1035084","url":null,"abstract":"This paper proposes an improved method for systematic decomposition of Boolean operators into a sequence of simpler ones. The improvement has two main components: (i) a sufficient condition for decreasing the number of possible child-templates during decomposition; (ii) pointing out the template element, the elimination of which results in the possibly maximum increment of the robustness value of the template. Examples are presented to demonstrate the effectiveness of the proposed method, whose advantages and limitations are also discussed.","PeriodicalId":387716,"journal":{"name":"Proceedings of the 2002 7th IEEE International Workshop on Cellular Neural Networks and Their Applications","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121185024","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 : 2002-07-22DOI: 10.1109/CNNA.2002.1035103
D. Bálya, T. Roska
Fast and robust classification of feature vectors is a crucial task in a number of real-time systems. A cellular neural/nonlinear network universal machine (CNN-UM) can be applied very efficiently as a feature detector and also for post-processing the results for object recognition. This paper shows how a robust classification scheme based on adaptive resonance theory (ART) can also be mapped to the CNN-UM. The designed analogic CNN algorithm is capable of classifying the extracted binary feature vectors keeping the advantages of the ART networks. An analogic algorithm is presented for unsupervised classification with tunable sensitivity and automatic new class creation. Another CNN-UM algorithm is suggested for supervised classification. In addition to the two algorithms, a new "repair" function is proposed to reduce the number of the created classes. The presented binary feature vector classification is feasible on the existing standard CNN-UM chips.
{"title":"Supervised and unsupervised art-like classifications of binary vectors on the CNN universal machine","authors":"D. Bálya, T. Roska","doi":"10.1109/CNNA.2002.1035103","DOIUrl":"https://doi.org/10.1109/CNNA.2002.1035103","url":null,"abstract":"Fast and robust classification of feature vectors is a crucial task in a number of real-time systems. A cellular neural/nonlinear network universal machine (CNN-UM) can be applied very efficiently as a feature detector and also for post-processing the results for object recognition. This paper shows how a robust classification scheme based on adaptive resonance theory (ART) can also be mapped to the CNN-UM. The designed analogic CNN algorithm is capable of classifying the extracted binary feature vectors keeping the advantages of the ART networks. An analogic algorithm is presented for unsupervised classification with tunable sensitivity and automatic new class creation. Another CNN-UM algorithm is suggested for supervised classification. In addition to the two algorithms, a new \"repair\" function is proposed to reduce the number of the created classes. The presented binary feature vector classification is feasible on the existing standard CNN-UM chips.","PeriodicalId":387716,"journal":{"name":"Proceedings of the 2002 7th IEEE International Workshop on Cellular Neural Networks and Their Applications","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130964355","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 : 2002-07-22DOI: 10.1109/CNNA.2002.1035083
L. Torok, Á. Zarándy
Color constancy (CC) is a perceptional phenomena in which living species, capable of color vision, perceive object color apart from the spectral distribution of light applied to illuminate the objects. The algorithm that can recover objects' original color and display them as if they were illuminated by spectrally even (white) light is called the CC algorithm. In contrast to other solutions our approach offers on-line possibilities in applications as its operation needs consist of mainly local interactions that are well suited to the architecture of cellular neural/non-linear networks (CNN). In a recent paper, we offered a brief survey of common CC approaches, introduced the principles of our CC algorithm, compared ACE4K on-chip results versus simulation, examined the robustness of our algorithm and outlined a newly developed setup for reliable color image recording.
{"title":"CNN based color constancy algorithm","authors":"L. Torok, Á. Zarándy","doi":"10.1109/CNNA.2002.1035083","DOIUrl":"https://doi.org/10.1109/CNNA.2002.1035083","url":null,"abstract":"Color constancy (CC) is a perceptional phenomena in which living species, capable of color vision, perceive object color apart from the spectral distribution of light applied to illuminate the objects. The algorithm that can recover objects' original color and display them as if they were illuminated by spectrally even (white) light is called the CC algorithm. In contrast to other solutions our approach offers on-line possibilities in applications as its operation needs consist of mainly local interactions that are well suited to the architecture of cellular neural/non-linear networks (CNN). In a recent paper, we offered a brief survey of common CC approaches, introduced the principles of our CC algorithm, compared ACE4K on-chip results versus simulation, examined the robustness of our algorithm and outlined a newly developed setup for reliable color image recording.","PeriodicalId":387716,"journal":{"name":"Proceedings of the 2002 7th IEEE International Workshop on Cellular Neural Networks and Their Applications","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129669389","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 : 2002-07-22DOI: 10.1109/CNNA.2002.1035091
R. Schonmeyer, D. Feiden, R. Tetzlaff
Cellular neural networks (CNN) are often considered as massive parallel computing arrays for high speed image processing. In order to find appropriate CNN templates, optimization methods are necessary in many cases. We consider the optimization method Iterative Annealing directly using the output of a hardware realization of a CNN-UM Chip. The procedure presented in this contribution generates highly adapted sets of templates for complex image processing tasks. With this approach it is also possible to tune existing CNN programs to compensate inaccuracies of analog CNN hardware leading to noise reduction and more robust behaviour. Finally, an application of practical interest has been developed, by using the introduced method. We achieved the tracing of a certain selected object out of an image sequence showing many moving objects.
{"title":"Multi-template training for image processing with cellular neural networks","authors":"R. Schonmeyer, D. Feiden, R. Tetzlaff","doi":"10.1109/CNNA.2002.1035091","DOIUrl":"https://doi.org/10.1109/CNNA.2002.1035091","url":null,"abstract":"Cellular neural networks (CNN) are often considered as massive parallel computing arrays for high speed image processing. In order to find appropriate CNN templates, optimization methods are necessary in many cases. We consider the optimization method Iterative Annealing directly using the output of a hardware realization of a CNN-UM Chip. The procedure presented in this contribution generates highly adapted sets of templates for complex image processing tasks. With this approach it is also possible to tune existing CNN programs to compensate inaccuracies of analog CNN hardware leading to noise reduction and more robust behaviour. Finally, an application of practical interest has been developed, by using the introduced method. We achieved the tracing of a certain selected object out of an image sequence showing many moving objects.","PeriodicalId":387716,"journal":{"name":"Proceedings of the 2002 7th IEEE International Workshop on Cellular Neural Networks and Their Applications","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122907974","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 : 2002-07-22DOI: 10.1109/CNNA.2002.1035044
D. Monnin, A. Koneke, J. Hérault
As soon as an image processing operator can be expressed as a linearly separable Boolean function involving a cell and its neighborhood, there is a way of straightforwardly deriving an equivalent cellular neural network (CNN) operation. An appropriate method had already been introduced for the robust design of uniformly initialized uncoupled CNN operators, and is now applied to the design of binary initialized and coupled CNN operators. A way of implementing in a unique operator two different Boolean functions conditioning the white-to-black and the black-to-white transitions, respectively, is also presented.
{"title":"Boolean design of binary initialized and coupled CNN image processing operators","authors":"D. Monnin, A. Koneke, J. Hérault","doi":"10.1109/CNNA.2002.1035044","DOIUrl":"https://doi.org/10.1109/CNNA.2002.1035044","url":null,"abstract":"As soon as an image processing operator can be expressed as a linearly separable Boolean function involving a cell and its neighborhood, there is a way of straightforwardly deriving an equivalent cellular neural network (CNN) operation. An appropriate method had already been introduced for the robust design of uniformly initialized uncoupled CNN operators, and is now applied to the design of binary initialized and coupled CNN operators. A way of implementing in a unique operator two different Boolean functions conditioning the white-to-black and the black-to-white transitions, respectively, is also presented.","PeriodicalId":387716,"journal":{"name":"Proceedings of the 2002 7th IEEE International Workshop on Cellular Neural Networks and Their Applications","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134237156","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 : 2002-07-22DOI: 10.1109/CNNA.2002.1035045
G. Liñán, Á. Rodríguez-Vázquez, S. Espejo, R. Domínguez-Castro
This paper presents a new generation 128/spl times/128 focal-plane analog programmable array processor (FPAPAP), from a system level perspective, which has been manufactured in a 0.35 /spl mu/m standard digital 1P-5M CMOS technology. The chip has been designed to achieve the high-speed and moderate-accuracy (8b) requirements of most real time early-vision processing applications. It is easily embedded in conventional digital hosting systems: external data interchange and control are completely digital. The chip contains close to four millions transistors, 90% of them working in analog mode, and exhibits a relatively low power consumption-<4 W, i.e. less than 1 /spl mu/W per transistor. Computing vs. power peak values are in the order of 1 TeraOPS/W, while maintained VGA processing throughputs of 100 frames/s are possible with about 10-20 basic image processing tasks on each frame.
{"title":"ACE16K: a 128/spl times/128 focal plane analog processor with digital I/O","authors":"G. Liñán, Á. Rodríguez-Vázquez, S. Espejo, R. Domínguez-Castro","doi":"10.1109/CNNA.2002.1035045","DOIUrl":"https://doi.org/10.1109/CNNA.2002.1035045","url":null,"abstract":"This paper presents a new generation 128/spl times/128 focal-plane analog programmable array processor (FPAPAP), from a system level perspective, which has been manufactured in a 0.35 /spl mu/m standard digital 1P-5M CMOS technology. The chip has been designed to achieve the high-speed and moderate-accuracy (8b) requirements of most real time early-vision processing applications. It is easily embedded in conventional digital hosting systems: external data interchange and control are completely digital. The chip contains close to four millions transistors, 90% of them working in analog mode, and exhibits a relatively low power consumption-<4 W, i.e. less than 1 /spl mu/W per transistor. Computing vs. power peak values are in the order of 1 TeraOPS/W, while maintained VGA processing throughputs of 100 frames/s are possible with about 10-20 basic image processing tasks on each frame.","PeriodicalId":387716,"journal":{"name":"Proceedings of the 2002 7th IEEE International Workshop on Cellular Neural Networks and Their Applications","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129299965","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 : 2002-07-22DOI: 10.1109/CNNA.2002.1035048
L. Orzó, S.T. Kes, T. Roska
A programmable opto-electronic analogic CNN computer (POAC) provides an efficient frame for diverse image processing applications, as it combines the enormous inherent computational capabilities of our new, massively parallel, but flexibly programmable optical CNN implementation with the capabilities of a visual CNN-UM chip. Our optical CNN implementation is based on an original, semi-incoherent optical correlator architecture, which is superior to other optical implementations in several respects. It makes real time reprogramming of a new type of joint Fourier transform correlator (t/sub 2/-JTC) possible while preserving the inherent speed of VanderLugt type of systems. Furthermore the POAC architecture overcomes the main limitations of both the microelectronic (VLSI) and other optical implementations. In this paper it will be shown that this device is particularly useful in image-processing algorithms, which cannot be fulfilled real time by any other existing optical or digital system due to the high number of pattern matching tasks required.
{"title":"Application issues of a programmable optical CNN implementation","authors":"L. Orzó, S.T. Kes, T. Roska","doi":"10.1109/CNNA.2002.1035048","DOIUrl":"https://doi.org/10.1109/CNNA.2002.1035048","url":null,"abstract":"A programmable opto-electronic analogic CNN computer (POAC) provides an efficient frame for diverse image processing applications, as it combines the enormous inherent computational capabilities of our new, massively parallel, but flexibly programmable optical CNN implementation with the capabilities of a visual CNN-UM chip. Our optical CNN implementation is based on an original, semi-incoherent optical correlator architecture, which is superior to other optical implementations in several respects. It makes real time reprogramming of a new type of joint Fourier transform correlator (t/sub 2/-JTC) possible while preserving the inherent speed of VanderLugt type of systems. Furthermore the POAC architecture overcomes the main limitations of both the microelectronic (VLSI) and other optical implementations. In this paper it will be shown that this device is particularly useful in image-processing algorithms, which cannot be fulfilled real time by any other existing optical or digital system due to the high number of pattern matching tasks required.","PeriodicalId":387716,"journal":{"name":"Proceedings of the 2002 7th IEEE International Workshop on Cellular Neural Networks and Their Applications","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131686226","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 : 2002-07-22DOI: 10.1109/CNNA.2002.1035055
W. Singer
Psychophysical and neurophysiological evidence indicates that the brain identifies perceptual objects by decomposing them into components by analyzing the relations among the respective components and representing in a combined code the components and their specific relations. This is an efficient strategy for two reasons. First, it permits unambiguous descriptions of a virtually unlimited number of perceptual objects with a limited set of symbols for components and relations. Second, it can be scaled and applied also for the description of complex constellations, i.e. for the infinite variety of contextual configurations in which perceptual objects can occur. Linguistic descriptions follow the same principle. By recombining in ever changing configurations a rather limited set of symbols for components, properties and relations, a virtually inexhaustible universe of constellations can be encoded. However, there is an interesting trade-off between the complexity of the symbols and the syntactic rules required for the definition of relations.
{"title":"Time as coding space for information processing in the cerebral cortex","authors":"W. Singer","doi":"10.1109/CNNA.2002.1035055","DOIUrl":"https://doi.org/10.1109/CNNA.2002.1035055","url":null,"abstract":"Psychophysical and neurophysiological evidence indicates that the brain identifies perceptual objects by decomposing them into components by analyzing the relations among the respective components and representing in a combined code the components and their specific relations. This is an efficient strategy for two reasons. First, it permits unambiguous descriptions of a virtually unlimited number of perceptual objects with a limited set of symbols for components and relations. Second, it can be scaled and applied also for the description of complex constellations, i.e. for the infinite variety of contextual configurations in which perceptual objects can occur. Linguistic descriptions follow the same principle. By recombining in ever changing configurations a rather limited set of symbols for components, properties and relations, a virtually inexhaustible universe of constellations can be encoded. However, there is an interesting trade-off between the complexity of the symbols and the syntactic rules required for the definition of relations.","PeriodicalId":387716,"journal":{"name":"Proceedings of the 2002 7th IEEE International Workshop on Cellular Neural Networks and Their Applications","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129293539","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}