US 12,136,286 B2
Method and system for keypoint extraction from images of documents
Kushagra Mahajan, Gurgaon (IN); Monika Sharma, Gurgaon (IN); and Lovekesh Vig, Gurgaon (IN)
Assigned to Tata Consultancy Services Limited, Mumbai (IN)
Appl. No. 17/607,437
Filed by Tata Consultancy Services Limited, Mumbai (IN)
PCT Filed Sep. 6, 2020, PCT No. PCT/IN2020/050778
§ 371(c)(1), (2) Date Oct. 29, 2021,
PCT Pub. No. WO2021/044447, PCT Pub. Date Mar. 11, 2021.
Claims priority of application No. 201921035983 (IN), filed on Sep. 6, 2019.
Prior Publication US 2022/0215683 A1, Jul. 7, 2022
Int. Cl. G06V 30/41 (2022.01); G06V 30/19 (2022.01)
CPC G06V 30/41 (2022.01) [G06V 30/19007 (2022.01)] 8 Claims
OG exemplary drawing
 
1. A processor implemented method for keypoint extraction, comprising:
collecting, via one or more hardware processors, a reference image and a test image of a document, as input, wherein the reference image is an image of a document template and the test image is an image of a used version of the document template; and
performing a keypoint extraction from the test image and the reference image, via the one or more hardware processors, comprising:
determining coordinates of bounding boxes for each of a plurality of words present in the reference image and the test image, by applying at least one Optical Character Recognition (OCR) technique on the reference image and the test image;
determining whether the plurality of words contain at least one character belonging to at least one of a plurality of pre-defined lists, based on at least one selection criteria, wherein each of the plurality of pre-defined lists comprises a plurality of characters characterized as having distinct tips at specific positions of character, wherein the plurality of pre-defined lists comprise a begCharList, an endCharList, a topCharList, and a bottom CharList;
extracting all words containing the at least one character belonging to at least one of a plurality of pre-defined lists, from the test image and the reference image, based on the at least one selection criteria;
identifying all connected components in each of the extracted words using an algorithm to find connected components;
determining whether keypoints are to be searched in a first component or in a last component of all the identified connected components, of each of the extracted words, based on the at least one selection criteria; and
searching in at least one of the first component or the last component and extracting at least four of the keypoints from the test image and corresponding at least four keypoints from the reference image.