Pub Date : 2006-06-07DOI: 10.1109/ICCIS.2006.252347
P. Park, S. Kim, J. Moon, M. Shin
This paper presents an efficient MPC algorithm for uncertain time-varying systems with input constraints. The proposed algorithm adopts the method of increasing free control horizon in the dual mode (i.e., a free control mode in the first finite horizon and a state-feedback mode in the following infinite horizon) paradigm so as to enlarge the set of stabilizable initial states. In the method, however, since the number of LMIs growing exponentially with the free control horizon makes the corresponding optimization problems intractable even for small horizon, it is impracticable to blindly increase the free control horizon. The objective of this paper is to relax the restriction on increase of the free control horizon, incurred on computational burdens in the method. By choosing a combination of hyper-boxes including a possible region of the initial states and then by designing a priori zone controller for each hyper-box so as to send any initial states in the hyper-box into the invariant ellipsoidal target set, the algorithm can dramatically reduce the on-line computational burden for enlarging the set of stabilizable initial states
{"title":"An Efficient MPC Algorithm based on a Priori Zone Control","authors":"P. Park, S. Kim, J. Moon, M. Shin","doi":"10.1109/ICCIS.2006.252347","DOIUrl":"https://doi.org/10.1109/ICCIS.2006.252347","url":null,"abstract":"This paper presents an efficient MPC algorithm for uncertain time-varying systems with input constraints. The proposed algorithm adopts the method of increasing free control horizon in the dual mode (i.e., a free control mode in the first finite horizon and a state-feedback mode in the following infinite horizon) paradigm so as to enlarge the set of stabilizable initial states. In the method, however, since the number of LMIs growing exponentially with the free control horizon makes the corresponding optimization problems intractable even for small horizon, it is impracticable to blindly increase the free control horizon. The objective of this paper is to relax the restriction on increase of the free control horizon, incurred on computational burdens in the method. By choosing a combination of hyper-boxes including a possible region of the initial states and then by designing a priori zone controller for each hyper-box so as to send any initial states in the hyper-box into the invariant ellipsoidal target set, the algorithm can dramatically reduce the on-line computational burden for enlarging the set of stabilizable initial states","PeriodicalId":296028,"journal":{"name":"2006 IEEE Conference on Cybernetics and Intelligent Systems","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133740552","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 : 2006-06-07DOI: 10.1109/ICCIS.2006.252245
Y. Yoshida
In this paper, we discuss an evaluation method of fuzzy numbers as mean values and measurement of fuzziness defined by fuzzy measures, and the presented method is applicable to fuzzy numbers and fuzzy stochastic process defined by fuzzy numbers/fuzzy random variables in decision making. We compare the measurement of fuzziness and the variance as a factor to measure uncertainty. Formulae are also given to apply the results to triangle-type fuzzy numbers and trapezoidal-type fuzzy numbers
{"title":"Mean Values of Fuzzy Numbers and the Measurement of Fuzziness by Evaluation Measures","authors":"Y. Yoshida","doi":"10.1109/ICCIS.2006.252245","DOIUrl":"https://doi.org/10.1109/ICCIS.2006.252245","url":null,"abstract":"In this paper, we discuss an evaluation method of fuzzy numbers as mean values and measurement of fuzziness defined by fuzzy measures, and the presented method is applicable to fuzzy numbers and fuzzy stochastic process defined by fuzzy numbers/fuzzy random variables in decision making. We compare the measurement of fuzziness and the variance as a factor to measure uncertainty. Formulae are also given to apply the results to triangle-type fuzzy numbers and trapezoidal-type fuzzy numbers","PeriodicalId":296028,"journal":{"name":"2006 IEEE Conference on Cybernetics and Intelligent Systems","volume":"60 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131395221","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 : 2006-06-07DOI: 10.1109/ICCIS.2006.252313
Photchanan Ratanajaipan, E. Nantajeewarawat, V. Wuwongse
An application profile specifies a set of terms, drawn from one or more standard namespaces, for annotation of data, and constrains their usage and interpretations in a particular local application. An approach to defining an application profile using the OWL and OWL/XDD languages is proposed - the former is a standard Web ontology language and the latter is a definite-clause-style knowledge representation language that uses XML expressions as their underlying data structure. Constraints are defined in terms of rules, which are represented as XDD clauses. As an illustration, application of the approach to defining Dublin core metadata initiative's library application profile (DC-Lib), along with the possibility of extending it by describing finer-grained semantic constraints, is demonstrated. A prototype catalog validation system has been implemented, and some experimental results are shown
{"title":"OWL/XDD: A Formal Language for Application Profiles","authors":"Photchanan Ratanajaipan, E. Nantajeewarawat, V. Wuwongse","doi":"10.1109/ICCIS.2006.252313","DOIUrl":"https://doi.org/10.1109/ICCIS.2006.252313","url":null,"abstract":"An application profile specifies a set of terms, drawn from one or more standard namespaces, for annotation of data, and constrains their usage and interpretations in a particular local application. An approach to defining an application profile using the OWL and OWL/XDD languages is proposed - the former is a standard Web ontology language and the latter is a definite-clause-style knowledge representation language that uses XML expressions as their underlying data structure. Constraints are defined in terms of rules, which are represented as XDD clauses. As an illustration, application of the approach to defining Dublin core metadata initiative's library application profile (DC-Lib), along with the possibility of extending it by describing finer-grained semantic constraints, is demonstrated. A prototype catalog validation system has been implemented, and some experimental results are shown","PeriodicalId":296028,"journal":{"name":"2006 IEEE Conference on Cybernetics and Intelligent Systems","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129824083","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 : 2006-06-07DOI: 10.1109/ICCIS.2006.252325
Behrang Q. Zadeh, A. Ilkhani, A. Ganjeii
This paper introduces an adaptive, language independent, and 'built-in error pattern free' spell checker. Proposed system suggests proper form of misspelled words using non deterministic traverse of 'ternary search tree' data structure. In other words the problem of spell checking is addressed by traverse a tree with variable weighted edges. The proposed system uses interaction with user to learn error pattern of media. In this way, system improves its suggestions as time goes by
{"title":"Adaptive Language Independent Spell Checking using Intelligent Traverse on a Tree","authors":"Behrang Q. Zadeh, A. Ilkhani, A. Ganjeii","doi":"10.1109/ICCIS.2006.252325","DOIUrl":"https://doi.org/10.1109/ICCIS.2006.252325","url":null,"abstract":"This paper introduces an adaptive, language independent, and 'built-in error pattern free' spell checker. Proposed system suggests proper form of misspelled words using non deterministic traverse of 'ternary search tree' data structure. In other words the problem of spell checking is addressed by traverse a tree with variable weighted edges. The proposed system uses interaction with user to learn error pattern of media. In this way, system improves its suggestions as time goes by","PeriodicalId":296028,"journal":{"name":"2006 IEEE Conference on Cybernetics and Intelligent Systems","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134634729","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 : 2006-06-07DOI: 10.1109/ICCIS.2006.252320
Ting-Cheng Chang, Chuen-Jiuan Jane, Yuan-Paio Lee
The main purpose of paper is to establish a system, which combines rough set and grey theory. This model is used to let the time-serial, season-serial or regular data have the dynamic trend concepts by grey prediction, then, select the data sets with trend value through rough set screening system. It mainly is applied for a portfolio prediction in the stock market. Our study first predicts each listed company's attributes of condition and decision-making by grey prediction, secondly groups their attributes by K-means grouping tools, then filters and categorizes the groups with the classified capacity of rough set for uncertain and non-sufficient information and selects the stock portfolio. And then we evaluate the company shares from the portfolio according to their past EPS and ROE and elect the better ones again. Finally, the selected companies are arranged in order with grey relation and determine the weight of each share in the portfolio according to it. The experimental result in Taiwan: during five years (2000-2004), the average annual rate of return was 38.1%. The portfolio determined by the model overran the market dramatically
{"title":"A Forecasting Model of Dynamic Grey Rough Set and its Application on Stock Selection","authors":"Ting-Cheng Chang, Chuen-Jiuan Jane, Yuan-Paio Lee","doi":"10.1109/ICCIS.2006.252320","DOIUrl":"https://doi.org/10.1109/ICCIS.2006.252320","url":null,"abstract":"The main purpose of paper is to establish a system, which combines rough set and grey theory. This model is used to let the time-serial, season-serial or regular data have the dynamic trend concepts by grey prediction, then, select the data sets with trend value through rough set screening system. It mainly is applied for a portfolio prediction in the stock market. Our study first predicts each listed company's attributes of condition and decision-making by grey prediction, secondly groups their attributes by K-means grouping tools, then filters and categorizes the groups with the classified capacity of rough set for uncertain and non-sufficient information and selects the stock portfolio. And then we evaluate the company shares from the portfolio according to their past EPS and ROE and elect the better ones again. Finally, the selected companies are arranged in order with grey relation and determine the weight of each share in the portfolio according to it. The experimental result in Taiwan: during five years (2000-2004), the average annual rate of return was 38.1%. The portfolio determined by the model overran the market dramatically","PeriodicalId":296028,"journal":{"name":"2006 IEEE Conference on Cybernetics and Intelligent Systems","volume":"48 11","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114060099","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 : 2006-06-07DOI: 10.1109/ICCIS.2006.252314
W. Wettayaprasit, U. Sangket
This paper presents a method of linguistic rule extraction from neural networks nodes pruning using frequency interval data representation. The method composes of two steps which are 1) neural networks nodes pruning by analysis on the maximum weight and 2) linguistic rule extraction using frequency interval data representation. The study has tested with the benchmark data sets such as heart disease, Wisconsin breast cancer, Pima Indians diabetes, and electrocardiography data set of heart disease patients from hospitals in Thailand. The study found that the linguistic rules received had high accuracy and easy to understand. The number of rules and the number of conjunction of conditions were small and the training time was also decreased
{"title":"Linguistic Knowledge Extraction from Neural Networks Using Maximum Weight and Frequency Data Representation","authors":"W. Wettayaprasit, U. Sangket","doi":"10.1109/ICCIS.2006.252314","DOIUrl":"https://doi.org/10.1109/ICCIS.2006.252314","url":null,"abstract":"This paper presents a method of linguistic rule extraction from neural networks nodes pruning using frequency interval data representation. The method composes of two steps which are 1) neural networks nodes pruning by analysis on the maximum weight and 2) linguistic rule extraction using frequency interval data representation. The study has tested with the benchmark data sets such as heart disease, Wisconsin breast cancer, Pima Indians diabetes, and electrocardiography data set of heart disease patients from hospitals in Thailand. The study found that the linguistic rules received had high accuracy and easy to understand. The number of rules and the number of conjunction of conditions were small and the training time was also decreased","PeriodicalId":296028,"journal":{"name":"2006 IEEE Conference on Cybernetics and Intelligent Systems","volume":"65 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124850371","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 : 2006-06-07DOI: 10.1109/ICCIS.2006.252276
W. Mo, S. Guan
This paper presents a new evolutionary approach for function optimization problems particle swarm assisted incremental evolution strategy (PIES). Two strategies are proposed. One is incremental optimization that the whole evolution consists of several phases and one more variable is focused in each phase. The number of phases is equal to the number of variables in maximum. Each phase is composed of two stages: in the single-variable evolution (SVE) stage, a population is evolved with respect to one independent variable in a series of cutting planes; in the multi-variable evolving (MVE) stage, the initial population is formed by integrating the population obtained by the SVE in current phase and by the MVE in the last phase. And then the MVE is taken on the incremented variable set. The second strategy is a hybrid of particle swarm optimization (PSO) and the evolution strategy (ES). PSO is applied to adjust the cutting planes (in SVEs) or hyper-planes (in MVEs) while ES is applied to searching optima in the cutting planes/hyper-planes. The results of experiments show that PIES generally outperforms three other evolutionary algorithms, improved normal GA, PSO and SADEXERAF, in the sense that PIES finds solutions with more optimal objective values and closer to the true optima
{"title":"Particle Swarm Assisted Incremental Evolution Strategy for Function Optimization","authors":"W. Mo, S. Guan","doi":"10.1109/ICCIS.2006.252276","DOIUrl":"https://doi.org/10.1109/ICCIS.2006.252276","url":null,"abstract":"This paper presents a new evolutionary approach for function optimization problems particle swarm assisted incremental evolution strategy (PIES). Two strategies are proposed. One is incremental optimization that the whole evolution consists of several phases and one more variable is focused in each phase. The number of phases is equal to the number of variables in maximum. Each phase is composed of two stages: in the single-variable evolution (SVE) stage, a population is evolved with respect to one independent variable in a series of cutting planes; in the multi-variable evolving (MVE) stage, the initial population is formed by integrating the population obtained by the SVE in current phase and by the MVE in the last phase. And then the MVE is taken on the incremented variable set. The second strategy is a hybrid of particle swarm optimization (PSO) and the evolution strategy (ES). PSO is applied to adjust the cutting planes (in SVEs) or hyper-planes (in MVEs) while ES is applied to searching optima in the cutting planes/hyper-planes. The results of experiments show that PIES generally outperforms three other evolutionary algorithms, improved normal GA, PSO and SADEXERAF, in the sense that PIES finds solutions with more optimal objective values and closer to the true optima","PeriodicalId":296028,"journal":{"name":"2006 IEEE Conference on Cybernetics and Intelligent Systems","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125145324","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 : 2006-06-07DOI: 10.1109/ICCIS.2006.252336
B. Chandra, P. Paul V
Decision trees algorithms have been suggested in the past for classification of numeric as well as categorical attributes. SLIQ algorithm was proposed (Mehta et al., 1996) as an improvement over ID3 and C4.5 algorithms (Quinlan, 1993). Elegant Decision Tree Algorithm was proposed (Chandra et al. 2002) to improve the performance of SLIQ. In this paper a novel approach has been presented for the choice of split value of attributes. The issue of reducing the number of split points has been addressed. It has been shown on various datasets taken from UCI machine learning data repository that this approach gives better classification accuracy as compared to C4.5, SLIQ and Elegant Decision Tree Algorithm (EDTA) and at the same time the number of split points to be evaluated is much less compared to that of SLIQ and EDTA
决策树算法在过去被建议用于数字和分类属性的分类。作为对ID3和C4.5算法的改进(Quinlan, 1993)，提出了SLIQ算法(Mehta et al.， 1996)。为了提高SLIQ的性能，提出了优雅决策树算法(Chandra et al. 2002)。本文提出了一种新的属性分割值选择方法。减少分界点数目的问题已得到解决。从UCI机器学习数据库中获取的各种数据集表明，与C4.5、SLIQ和优雅决策树算法(EDTA)相比，这种方法具有更好的分类精度，同时与SLIQ和EDTA相比，需要评估的分裂点数量要少得多
{"title":"A Robust Algorithm for Classification Using Decision Trees","authors":"B. Chandra, P. Paul V","doi":"10.1109/ICCIS.2006.252336","DOIUrl":"https://doi.org/10.1109/ICCIS.2006.252336","url":null,"abstract":"Decision trees algorithms have been suggested in the past for classification of numeric as well as categorical attributes. SLIQ algorithm was proposed (Mehta et al., 1996) as an improvement over ID3 and C4.5 algorithms (Quinlan, 1993). Elegant Decision Tree Algorithm was proposed (Chandra et al. 2002) to improve the performance of SLIQ. In this paper a novel approach has been presented for the choice of split value of attributes. The issue of reducing the number of split points has been addressed. It has been shown on various datasets taken from UCI machine learning data repository that this approach gives better classification accuracy as compared to C4.5, SLIQ and Elegant Decision Tree Algorithm (EDTA) and at the same time the number of split points to be evaluated is much less compared to that of SLIQ and EDTA","PeriodicalId":296028,"journal":{"name":"2006 IEEE Conference on Cybernetics and Intelligent Systems","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125810324","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 : 2006-06-07DOI: 10.1109/ICCIS.2006.252292
N. Umashankar, V. Karthik
In flexible manufacturing systems (FMS), automated guided vehicles (AGVs) are used for transportation of the processed materials between various pickup and delivery points. The assignment of an AGV to a workcentre from a set of workcentres simultaneously requesting the service for transport of a part is often solved in real-time with simple dispatching rules. This paper proposes an intelligent dispatching approach for the AGVs based on multi-criteria fuzzy logic controller, which simultaneously takes into account multiple aspects in every dispatching decision. The controller operates in two stages in which the second stage is constructed as a conflict resolving tool between two equally ranked AGVs for a particular workcentre. The control system is being implemented using MATLAB and its fuzzy inference engine. Sample runs have been provided to illustrate the controller implementation
{"title":"Multi-criteria Intelligent Dispatching Control of Automated Guided Vehicles in FMS","authors":"N. Umashankar, V. Karthik","doi":"10.1109/ICCIS.2006.252292","DOIUrl":"https://doi.org/10.1109/ICCIS.2006.252292","url":null,"abstract":"In flexible manufacturing systems (FMS), automated guided vehicles (AGVs) are used for transportation of the processed materials between various pickup and delivery points. The assignment of an AGV to a workcentre from a set of workcentres simultaneously requesting the service for transport of a part is often solved in real-time with simple dispatching rules. This paper proposes an intelligent dispatching approach for the AGVs based on multi-criteria fuzzy logic controller, which simultaneously takes into account multiple aspects in every dispatching decision. The controller operates in two stages in which the second stage is constructed as a conflict resolving tool between two equally ranked AGVs for a particular workcentre. The control system is being implemented using MATLAB and its fuzzy inference engine. Sample runs have been provided to illustrate the controller implementation","PeriodicalId":296028,"journal":{"name":"2006 IEEE Conference on Cybernetics and Intelligent Systems","volume":"363 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126808497","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 : 2006-06-07DOI: 10.1109/ICCIS.2006.252285
M. Stojmenovic
Our primary interest is to build fast and reliable object recognizers in images based on small training sets. This is important in cases where the training set needs to be built mostly manually, as in the case that we studied, the recognition of the Honda Accord 2004 from rear views. We described a novel variant of the AdaBoost based learning algorithm, which builds a strong classifier by incremental addition of weak classifiers (WCs) that minimize the combined error of the already selected WCs. Each WC is trained only once, and examples do not change their weights. We proposed to pre-eliminate features whose cumulative error of corresponding best WCs exceeds a predetermined threshold value. We tested two straightforward definitions of cumulative error. In both cases, we showed that, when over 97% of the initial features are eliminated at the very beginning from further training, training time is drastically reduced while having little impact on the quality of the pool of available WCs. This is a novel method that has reduced the training set WC quantity to less than 3% of its original number, greatly speeding up training time, and showing no negative impact on the quality of the final classifier. Our experiments indicated that the set of features used by Viola and Jones and others for face recognition was inefficient for our problem; therefore, each object requires its own custom-made set of features for real time and accurate recognition. Our training method, combined with the selection of appropriate features, has resulted in finding a very accurate classifier containing merely 30 weak classifiers. Compared to existing literature, we have overall achieved the design of a real time object detection machine with the least number of examples, the least number of weak classifiers, the fastest training time, and with competitive detection and false positive rates
{"title":"Pre-Eliminating Features for Fast Training in Real Time Object Detection in Images with a Novel Variant of AdaBoost","authors":"M. Stojmenovic","doi":"10.1109/ICCIS.2006.252285","DOIUrl":"https://doi.org/10.1109/ICCIS.2006.252285","url":null,"abstract":"Our primary interest is to build fast and reliable object recognizers in images based on small training sets. This is important in cases where the training set needs to be built mostly manually, as in the case that we studied, the recognition of the Honda Accord 2004 from rear views. We described a novel variant of the AdaBoost based learning algorithm, which builds a strong classifier by incremental addition of weak classifiers (WCs) that minimize the combined error of the already selected WCs. Each WC is trained only once, and examples do not change their weights. We proposed to pre-eliminate features whose cumulative error of corresponding best WCs exceeds a predetermined threshold value. We tested two straightforward definitions of cumulative error. In both cases, we showed that, when over 97% of the initial features are eliminated at the very beginning from further training, training time is drastically reduced while having little impact on the quality of the pool of available WCs. This is a novel method that has reduced the training set WC quantity to less than 3% of its original number, greatly speeding up training time, and showing no negative impact on the quality of the final classifier. Our experiments indicated that the set of features used by Viola and Jones and others for face recognition was inefficient for our problem; therefore, each object requires its own custom-made set of features for real time and accurate recognition. Our training method, combined with the selection of appropriate features, has resulted in finding a very accurate classifier containing merely 30 weak classifiers. Compared to existing literature, we have overall achieved the design of a real time object detection machine with the least number of examples, the least number of weak classifiers, the fastest training time, and with competitive detection and false positive rates","PeriodicalId":296028,"journal":{"name":"2006 IEEE Conference on Cybernetics and Intelligent Systems","volume":"28 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2006-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116314138","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：