A New Strategy of OCR Combination

摘要

In this paper, we present a new method for combining OCR systems in order to optimize document recognition. The proposed method combines the results of individual OCR systems, which are classified into several categories according to their characteristics. A set of OCR systems are evaluated using several performance measures. The contribution of this paper is twofold. First of all, we assessed OCR system according to two new criteria: their performance in table recognition and their capacity to recognize graphs. Furthermore we defined a new strategy of OCR combination. Our hybrid strategy is based on three techniques: the image analysis, the parallel combination approach, and the majority vote method. None of the related combination methods was interested in the analysis of the document image. With this new method, we aim to increase the document recognition rate regardless to their structure (simple or complex) and it composition (text, graphs, tables, images, etc).