Automatic Script and Type Identification in Bi-lingual Forms

In this paper we have developed a system that can automatically discriminate between machine-printed and handwritten words in structured bi-lingual (Arabic and French) form document layout. Our system has been applied in the context of Tunisian National Health Insurance Fund for medical care costs refund with encouraging results. In the used forms, handwritten data usually touch or cross the preprinted form frames and texts, creating complex problems for the recognition routines. Each text type should also be processed using different methods in order to optimize the recognition accuracy. This work aims to address these issues and to especially solve the problem of machine-printed/handwritten and Arabic/French word discrimination. To this end, we computed co-occurrence matrix of oriented gradients from word’s image and used it as input to a k-Nearest Neighbor classifier. Experiments are carried on 20 forms. An average script identification rate of 98.31% is achieved.