Fuzzy nearest neighbor system: An application to Handwritten Arabic literal amount words recognition. Nadir FARAH, Mokhtar SELLAMI Laboratoire de recherche en informatique Université Badji Mokhtar BP 12 Annaba ALGERIE Abstract: In recent years, fuzzy logic has been increasingly used to improve conventional methods especially in pattern recognition fields. The aim of this paper is Arabic literal words amount recognition using a fuzzy classifier. We introduce briefly the technique for processing handwritten words, which begins with the extraction of features, then their classification. The purpose of the classifier is to allocate a class to the test word on a basis of a training set. The fuzzification is introduced in two stages, firstly to reclassify the obtained K nearest neighbors by a classical K nearest neighbors approach. Secondly in the classification of the tested word to a class among its K neighbors. The proposed system was tested with a wide range of test images and an interesting success rate of classification was obtained. Key Words: Word recognition, fuzzy nearest neighbors, membership value. 1. Introduction Handwritten word recognition is among the most widely studied fields. It supports not only statistical, structural information and semantic ones, but also some physiological and psychological state of the writer. This characteristic makes handwritten word recognition of consent especially in bank checks area. Word recognition has become during the later decades almost universal. From that, many automatic systems have been developed and implemented. Most existing systems deal with some constraints and the results are interesting. For example, a limited lexicon or restricted writer number. However, the handwritten deal with variability of the script and the noises generated by scanner. To recognize a word, a fuzzy K nearest neighbor [1] is implemented in the Arabic handwritten literal amount recognition system described in this paper. The proposed system we deal with consists of five parts, among them: data acquisition, preprocessing, feature extraction, recognition and post classification. In data acquisition a handwritten literal amount are captured by a scanner, after which preprocessing techniques are used to prepare the image of words for feature extraction. The preprocessing stage begins by dividing the literal amount into words,