Combined Document/Business Card Detector for Proactive Document-Based Services on the Smartphone

In this paper, we present a novel combined detector of document and business card. To detect document or business card, our method firstly extracts a document object region from a given image, and then classifies it into positive or negative class. In the step of extracting the document object region, a block-based processing is exploited to efficiently find the line segment candidates of its boundary, and RANSAC-like method under three constraints is used to search its real boundary. In classification step, after performing image normalization on the extracted region, the Fisher vector is extracted to represent the document object, then it is classified by linear-SVM. For evaluating the proposed method, we carry out some experiments by using the collected images, and show that our method has achieved about 94% accuracy.