An expert system for analysis and recognition of general symbols is introduced. The system uses the structural pattern recognition technique for modeling symbols by a set of straight lines referred to as segments. The system rotates, scales and thins the symbol, then extracts the symbol strokes. Each stroke is transferred into segments (straight lines). The system is shown to be able to map similar styles of the symbol to the same representation. When the system had some stored models for each symbol (an average of 97 models/symbol), the rejection rate was 16.1 % and the recognition rate was 83.9 % of which 95 % was recognized correctly. The system is tested by 5726 handwritten characters from the Center of Excellence for Document Analysis and Recognition (CEDAR) database. The system is capable of learning new symbols by simply adding their models to the system knowledge base. ( 2000 Pattern Recognition Society. Published

In this paper, the cooperation of two feature families for handwritten digit recognition using a committee of Neural Network (NN) classifiers will be examined. Various cooperation schemes will be investigated and corresponding results will be presented. To improve the system reliability,we will upgrade the committee scheme using multistage classification based on rule-based and statistical cooperation. The rule-based cooperation enables an easy and efficient implementation of various rejection criteria while the statistical cooperation offers better possibility for fine-tuning of the recognition versus the reliability tradeoff. The final system has been implemented using rule-based reasoning with rejection criteria for classifier decision fusion and the generalized committee cooperation scheme for classification of the rejected digit patterns. The presented results show that we propose a successful approach for reliability control in committee classifier environment and indicate that a suitable cooperation of statistical and rule-based decision fusion is a promising approach in handwritten recognition systems.

Abstract – In this paper, various cooperation schemes of SVM (Support Vector Machine) classifiers applied on two feature sets for handwritten digit recognition are examined. We start with a feature set composed of structural and statistical features and corresponding SVM classifier applied on the complete feature set. Later, we investigate the various partitions of the feature set as well as the advantages and weaknesses of various decision fusion schemes applied on SVM classifiers designed for partitioned feature sets. The obtained results show that it is difficult to exceed the recognition rate of a single SVM classifier applied straightforwardly on the complete feature set. Additionally, we show that the partitioning of the feature set according to feature nature (structural and statistical features) is not always the best way for designing classifier cooperation schemes. These results impose need of special feature selection procedures for optimal partitioning of the feature set for classifier cooperation schemes. Index terms – classification, committee, features, rejection, reliability 1.

this paper, we will talk only about OCR applications: while our approach is general to image processing, OCR serves as a perfect test bed for new algorithms, because of the extensive litterature and "universal" data bases (the NIST data base) available. and also of how big the learning set will have to be (i.e. how many data will be needed to match the unknown variables). Thus reducing the complexity of a NN is a means to reduce costs, and also, as we just saw, to increase performances: really a very good deal indeed ! As a consequence of this, a lot of work has been devoted, along the years, to designing networks with carefully controlled complexity. In particular, Time Delay Neural Networks-TDNN- [24], have been used to great success for OCR [12]: a TDNN is an architecture where complexity is controlled through connectivity (connections are local) and weights (which are shared). This dramatically reduces the TDNN complexity as compared to a fully connected architecture: for example the TDNN in [12] has 100 000 connections but only 2 500 weights, while a fully connected NN with 100 000 connections would have that many weights! Finding the optimal architecture though is very much of an art: there does not seem to exist any systematic approach to design the appropriate pattern of connections, local or shared weights. Basically, one tests, by-trial and-error, various choices for receptive field sizes and overlaps), and finally selects that architecture which yields the best performances on a validation set. This process is, of course, extremely time consuming and one could always fear that the final architecture is not optimal, but only the best one among those tested. This problem is often considered as a major weakness of#the NN approach. We need a principled way to d...

haci ôeef ~ ngcioduced from tho microfilm master. UMI films tha text directly hom thb original or copy submitted. Thur, mme thesis and dissertation copies am in typewriter ha, Whik dhem may be lrom any type of amputer printer. Th. qwllZy of this npmduction is dopendant upon li. qurlity of üm copy submitted. Broken or indistinct print, coloreâ or poor quality illustrations and photographs, print bbdthiwgh, substandaid margins, and improper alignmnt can advendy nfect nprodudion. In the unliWy event tht #a author diâ not rend UMI a compleb manucaipt and thers are missing pages. tbse will ba noW. Ab, if unouthorited copyright material had to be removed, a nole will indicote the deletion. Oversize materials (e.g., mapr, dnwings, charts) are nproduced by sectiming the originil, beginning at upper lefthsnd corner wd contiming from Wt to ngM in equal sdons with small werlapr. Photographs induded in üw original muscript have ôwn repioduœd xemgmphically in this m. Highw qurlity 6 ' x I Mck and white photognphic prînts are availaMe (br n y photogmphs or illustrations appearing in this copy for an additional charge. Contact UMI dimcüy to ooidrr. 8811 & Howdl Information a d L.wning

In this paper, various cooperation schemes of SVM (Support Vector Machine) classifiers applied on two feature sets for handwritten digit recognition are examined. We start with a feature set composed of structural and statistical features and corresponding SVM classifier applied on the complete feature set. Later, we investigate the various partitions of the feature set as well as the advantages and weaknesses of various decision fusion schemes applied on SVM classifiers designed for partitioned feature sets. The obtained results show that it is difficult to exceed the recognition rate of a single SVM classifier applied straightforwardly on the complete feature set. Additionally, we show that the partitioning of the feature set according to feature nature (structural and statistical features) is not always the best way for designing classifier cooperation schemes. These results impose need of special feature selection procedures for optimal partitioning of the feature set for classifier cooperation schemes.