Published in the Proceedings of the Third International Conference on Document Anal- ysis and Recognition, Montreal, pp.841-844,1995. Extracting Meaningful Handwriting Features with Fuzzy Aggregation Method Ashutosh Malaviya and Liliane Peters GMD - German National Research Center for Information Technology Schloß Birlinghoven, 53757 St. Augustin, Germany Ashutosh.Malaviya@gmd.de Abstract Recognition methods use different features to assign a pat- tern to a prototype class. The recognition accuracy strongly depends on the selected features. In this paper we present a novel fuzzy methodology to extract appropriate fuzzy features from the handwriting data. From these meaningful features a set of linguistic rules are derived which in turn constitute a fuzzy rule base for character rec- ognition. The fuzzy features are confined to their meaning- fulness with the help of multistage feature aggregation scheme. 1.0 Introduction A globally applicable handwriting recognition system should classify symbols written by different persons from various backgrounds e.g. nationality, education or profes- sion. Development of such a system is only possible if the character prototypes are defined with a flexible rule base including only handwriting style independent features. The primary goal of our methodology is to represent the hand- writing information into a minimum number of meaningful features. These features are subsequently stored into a com- pact data-base. The proposed method achieves the flexibil- ity in recognizing varied handwritings by using linguistic fuzzy rules. The human visual sense is selectively activated in response to curved lines and other geometrical characteristics [2]. These features themselves contain certain vagueness in terms of their definition. The human recognition system is most accurate in grasping the typical geometrical features of handwritten characters while ignoring the vagueness. To compensate this inherent geometrical shape distortions existing in the different handwriting styles we have utilized fuzzy set theory[8]. By means of it we integrate the exist- ing vagueness into membership function of basic structural features. These possibilities are then estimated and further processed with fuzzy aggregation techniques. The presented paper is structured as follows: First we give a brief definition of the used fuzzy aggregation operators. Then we describe the proposed aggregation algorithm that extracts the meaningful fuzzy features from the input data, this is illustrated by an example. In the fourth section experimental results are presented. 2.0 Theoretical background of fuzzy aggregation The main objective of the fuzzy aggregation mechanism is to find an overall measure of certain fuzzy information from an uncertain and imprecise source of distributed information data. A generalized fuzzy membership function which com- bines these distributed information data can be formulated as: (1) represents the aggregated fuzzy information in an uni- verse of infinite discourse [0,1]. In other words it repre- sents the degree of membership or quality of the overall information in the given context. The form of F is depen- dent upon the semantic and syntactic relations between these distributed informations (see Sec. 3). The fuzzy set theory provides a large number of aggrega- tion connectives[1][3]. In our approach we considered the Yager union connective and the weighted generalized means connective(WGM) aggregation operators. A detailed description of these operators an their properties μ C μ 1 μ 2 …μ m , , , ( ) μ C F μ 1 μ 2 …μ m , , , ( ) = μ C