Approach for Arabic Handwritten Image Processing: Case of Text Detection in Degraded Documents

This study presents a new approach for processing of Arabic handwritten documents based on the extraction of characteristics and mechanisms involved in the process of human visual perception. The architecture which has been developed is based on the concept of multi-agent systems, allowing the integration of different stages of character recognition process in a cooperative way. This is illustrated using as example the prepossessing of binary noisy document. Therefore, a method was proposed, in order to distinguish between text and non-text components, using a new geometric primitives extracted from the analysis of the characteristics of Arabic script. Results show pixel-level precision and recall respectively of 98% and 93% for noise removal. This proves the effectiveness of the proposed approach in processing degraded documents and, consequently, improving the recognition performance.