Invoice Content table analysis with feature fusion

Invoice processing and financial information extraction have been popular topics among researchers for decades. Corporations like participation banks process invoices manually. Extraction of important information from invoices like product, price, amount etc. is a prerequisite for these banks. In this paper we propose a novel technique for processing invoice image tables automatically. Invoice images we process are mostly sent by customers as scanned or fax images which have low image quality. In order to process these invoices we run different methods several times with different parameters. Results from each method are fused to get candidate tables. The proposed methods are robust to the character set used in a document, the image resolution and the noise ratio of the document image, and can perform detection operations in a highly effective manner. In addition to success in low quality images, this method can be applied both on tables with and without borders. The quantitative results obtained by applying this method on real business invoices have very favorable results.