econfigurable Printed Character Recognition System Using a Logic Synthesis Tool Henry Selvaraj and Muthukumar Venkatesan Gippsland School of Computing and Information Technology Monash University, Churchill - 3842, Australia. { Henry.Selvaraj, Venkatesan.Muthukumar) @fcit.monash.edu.au Abstract In recent years functional decomposition methods, widely known to logic synthesis researchers are being applied in diverse fields such as Machine Learning [ 141 [ 171, Knowledge Discovery [ 161 [ 171, Information Systems [ 51[7] [ 131 and Image Compression [lo]. This paper presents a novel method for recognising machine printed characters and character images using functional decomposition. The methods found in literature [9][20] try to find some characteristics of an image and apply different computational intelligence techniques to match them to a character. Each character or image is viewed as set of conditions with a corresponding set of decisions. This paper shows functional decomposition [4][6] as a tool in determining the characters by considering less number of conditions. Moreover, the structures produced by functional decomposition are easily implementable by FPGAs and therefore can be quickly reconfigured to suit a different set of characters. The problem of character recognition is analogous to decision making in information systems. The decision table (DT) generated from a set of characters is functionally decomposed and intermediate decision rules are generated. The advantage of the proposed method is in the analysis of the character recognition process using the intermediate conditions and decisions. 1. Introduction Character recognition is a process of converting scanned images of machine printed or handwritten text, comprising numerals, letters and other symbols, into computer processable formats. This work deals with the recognition of printed Roman characters and character images using functional decomposition. A character recognition system can be broadly divided into the following four major parts: 1089-6503/98 $10.00 0 1998 IEEE 1. Preprocessing 2. Self Learning 3. Recognition 4. Postprocessing There are two major methodologies used in the recognition part: 1 .Global Analysis Method 2. Structure Analysis Method During recognition, the features of input characters are extracted (feature extraction) and compared with features in the database. If the features match or partially match, the input characters are grouped into a class (classification). The characters in a class have some common features among them. The main purpose of feature extraction is to eliminate the redundant information, which are irrelevant to the attributes of the character. Functional decomposition is a process by which a function of n input variables and m output variables is re- expressed as a function of functions of fewer variables [ 11. The main strategies of decomposition that do not assume special type of blocks [4] are serial decomposition, parallel decomposition and generalised decomposition [3]. It is possible to derive serial and parallel decompositions from generalised decomposition. By iteratively applying the above decomposition strategies, a complex and a large function with large number of input variables, n (30 S n loo), can be represented as functions with smaller input variables. Functional decomposition in character recognition is analogous to classical classification methods. Functional decomposition of the attributes of the characters results in a multi-level structure, where each level contains rules for classification and recognition of characters. 24