عملکرد مدل مخفی مارکوف و کلاس بندی شبکه های بیزی پویا در شناخت کلمه دست نوشته های عربی

This paper presents a comparative study of two machine learning techniques for recognizing handwritten Arabic words, where hidden Markov models (HMMs) and dynamic Bayesian networks (DBNs) were evaluated. The work proposed is divided into three stages, namely preprocessing, feature extraction and classification. Preprocessing includes baseline estimation and normalization as well as segmentation. In the second stage, features are extracted from each of the normalized words, where a set of new features for handwritten Arabic words is proposed, based on a sliding window approach moving across the mirrored word image. The third stage is for classification and recognition, where machine learning is applied using HMMs and DBNs. In order to validate the techniques, extensive experiments were conducted using the IFN/ENIT database which contains 32,492 Arabic words. Experimental results and quantitative evaluations showed that HMM outperforms DBN in terms of higher recognition rate and lower complexity.

Handwriting recognition (HWR) is a mechanism for transforming the written text into a symbolic representation, which plays an essential role in many human–computer interaction applications including cheque verification, automatic mail sorting, office automation as well as natural human–computer interaction [1]. HWR for Latin languages has been conducted and significant achievements have been made. However, there has been less work in Arabic handwriting recognition. This is due to the complexity of the Arabic language and lack of public Arabic handwriting databases. In general, HWR can be categorized into two distinct types: online and off-line based systems. Recognition in online systems uses the dynamics of writing by following the pen movement. Recognition in off-line based systems is based solely on an image of the written text. Online recognition is easier because it can make use of the additional information not available to the off-line systems such as the strength and sequential order of the writing [2]. However, online recognition is not possible in many applications so in this paper, we focus on the off-line recognition of handwritten Arabic text. The recognition of handwritten Arabic scripts can be divided into segmentation based or segmentation free approaches. The former segments words into characters or letters for recognition and can be regarded as an analytical approach. The latter, which can be regarded as a global approach, takes the whole word image for recognition and therefore needs no segmentation. Although the global approach makes the recognition process simpler, it requires a larger input vocabulary than analytical approach [3]. This paper focuses on the Arabic handwritten word recognition phase and introduces new methods for extracting features. Several experiments have been conducted using the IFN/ENIT benchmark database [4] and our algorithm showed the best recognition rate among the existing work reported using the same database. The remainder of this paper is structured as follows: Section 2 presents the literature review while Sections 3 and 4 describe the proposed method in terms of pre-processing and feature extraction; Section 5 describes the HMM classification process in details; experimental results are presented in Section 6. The paper ends with conclusions and suggestions for further work.

In this paper, the performance of both HMMs and DBNs classifiers were compared in terms of recognizing the handwritten Arabic words. Both classifiers are superior in classifying handwritten and printed scripts. The performance of both HMMs and DBNs classifiers on handwritten Arabic word recognition are reported. Actually, HMM is an excellent in classifying Arabic handwritten words. The system has been applied to the well-known IFN/ENIT database containing handwriting words written by different writers. We have found that the pixel density features are effective in our classifiers, and good results of recognition rate have been achieved. In addition, this system can be applied to other patterns for recognition with slightly adaptation. The result obtained in this research show that the best performances are always reached by the HMMs. This shows the superiority of the HMMs over all various classifiers used. Regarding the speed, HMMs were faster in training and testing time.