Off-Line Roman Cursive Handwriting Recognition

Abstract

Automatic handwriting recognition has been a subject of research for more than 40 years [84]. On the one hand, the reading of human handwriting by machine has been considered an interesting and intellectually challenging problem in its own right. To approach, or even surpass, the performance of humans in text recognition has been a major driving force behind many research activities. On the other hand, the field has been quite important from the commercial and application-oriented point of view. Automatic address reading [88], bank cheque processing [43], and recognition of text filled in by hand on forms [25, 102, 105] have been major challenges in automatic handwriting recognition research. Moreover, handwritten data have often been used to validate and test the performance of new pattern classification methods. Since the beginning of the 1990s, a significant growth of activities in handwriting recognition research has been observed. There is no doubt that enormous progress has taken place in this area. For example, for the tasks of handwritten address reading and amount recognition on bank cheques, commercial systems have become available [26]. Nevertheless, there is a clear need to further develop the field. All successful applications, for example , address and cheque reading, work in narrow domains with limited vocabularies, where task-specific knowledge and constraints are available. Examples are the relation between zip code and city name in address reading , or the redundancy of courtesy and legal amount on a cheque. However, when it comes to general word or sentence recognition where no constraints exist and one is faced with a large, possibly open lexicon, the state of the art is quite limited and recognition rates are rather low. Yet the problem of unconstrained word and sentence recognition is important in a number of