Schema recognition using modified weighted direction index histogram for medical document retrieval systems

This paper discusses a schema recognition method considering annotations to improve recognition accuracy of the previous method. Previously, we studied document image recognition methods for medical/clinical documents and their applications. In addition, we also discussed the method to recognize schemas in the document, because they had important key information for medical document retrieval. However, annotations made a feature vector change drastically, as a result, reduction of the recognition accuracy was caused. In this paper, annotations were automatically detected, and the feature vectorconsidering them was generated. We conducted evaluation experiments using actual schemas used in Mie University Hospital. After experiments, the recognition accuracy of the proposed method was 96%.