2019/11/13 by Kushagra Mahajan, Monika Sharma, Mahajan, Kushagra +3 · 2 citations
Computer Science · #Advanced Image and Video Retrieval Techniques #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Handwritten Text Recognition Techniques #Image Retrieval and Classification Techniques
paper · pdf · doi:10.48550/arxiv.1911.05870
openalex publication_date 2019/11/13 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Precise homography estimation between multiple images is a pre-requisite for\nmany computer vision applications. One application that is particularly\nrelevant in today's digital era is the alignment of scanned or camera-captured\ndocument images such as insurance claim forms for information extraction.\nTraditional learning based approaches perform poorly due to the absence of an\nappropriate gradient. Feature based keypoint extraction techniques for\nhomography estimation in real scene images either detect an extremely large\nnumber of inconsistent keypoints due to sharp textual edges, or produce\ninaccurate keypoint correspondences due to variations in illumination and\nviewpoint differences between document images. In this paper, we propose a\nnovel algorithm for aligning scanned or camera-captured document images using\ncharacter based keypoints and a reference template. The algorithm is both fast\nand accurate and utilizes a standard Optical character recognition (OCR) engine\nsuch as Tesseract to find character based unambiguous keypoints, which are\nutilized to identify precise keypoint correspondences between two images.\nFinally, the keypoints are used to compute the homography mapping between a\ntest document and a template. We evaluated the proposed approach for\ninformation extraction on two real world anonymized datasets comprised of\nhealth insurance claim forms and the results support the viability of the\nproposed technique.\n