Toward Distributed Cursive Writing OCR Systems Based on a Combination of Complementary Approaches

Large amounts of cursive writing documents are still waiting to be computerized for several and different purposes. These documents are in general of medium to low quality; hence they require a sophisticated recognition algorithm capable of properly extracting the correct text from low quality cursive documents. The Dynamic Time Warp (DTW) algorithm is among the most effective algorithms for cursive writing optical character recognition (OCR). However, the DTW is a rather complex task requiring extensive computational capabilities, which hinders its commercial deployment on nonspecialized stand alone machines. Volunteer grids, such as XtremWeb and BOINC, provide viable infrastructures to speed up the DTW execution time. Recent experiments conducted on the Scientific Research Tunisian Grid (SRTG), an XtremWeb volunteer grid, confirmed this claim and showed a very tangible speedup along very high recognition rates. Such infrastructures present several practical advantages, such as the possibility of noncondemnation of the involved computers and the possibility of their simultaneous use by different users and/or applications. Unfortunately, volunteer grid infrastructures are inherently unable to guarantee the continuous availability of the stored data and, more importantly, the engaged processing capacities. Any involved computer may renegade and depart from the system at will, which consequently affects the application performance. Agent technology can be exploited here to solve the problem. In this chapter, we propose a service-oriented grid architecture (SOGA) based on the integration of both grid and agent technologies. An analytical study is conducted to ascertain and evaluate the key performance parameters of our proposed SOGA. The results confirm that our proposal provides a solid and viable solution for the large scale recognition of printed cursive writing based on the DTW algorithm.