公开(公告)号：GB2604970A

公开(公告)日：2022-09-21

申请号：GB202116057

申请日：2021-11-09

Applicant: IBM

Inventor： ZHONG FANG YUAN ,  YU PAN ,  TONG LIU ,  YI CHEN ZONG ,  LI JUAN GAO ,  QIONG WU

IPC: G06V30/164

Abstract: An optical character recognition (OCR) segmentation and processing method comprising: receiving a document 502; detecting different types of text data in the document 502a-d; dividing the document into a plurality of text regions containing a single type of text 510, 514a-d; removing optical noise from each text region 512, (600, Fig. 6); selecting a suitable OCR recognition model for each text region (704, 708, Fig. 7) to extracting the text (710, Fig. 7). The types of text detected may be tables, watermarks, handwriting and rotated text. Removing optical noise, such as unnecessary background text and/or images, from each text region may comprise: encoding each text region as a total semantic vector; dividing each text region into multiple sub-regions and encoding the sub-regions as a vector; generating a dot product for each sub-region; comparing the dot-product to a threshold; and deleting all pixels of each subregion that exceeds the threshold. Selection of the OCR code may be based on a classification of the text in a region (704, Fig. 7). The optical character recognition modules may comprise self-learning software such as neural networks.

公开(公告)号：GB2597166A

公开(公告)日：2022-01-19

申请号：GB202115690

申请日：2020-04-02

Applicant: IBM

Inventor： ZHONG FANG YUAN ,  KUN YAN YIN ,  YUAN LIN YANG ,  TONG LIU ,  HE LI

IPC: G06F16/00

Abstract: A set of candidate intent vectors is generated from an input intent vector. A validation of the set of candidate intent vectors is performed that selects as valid intent vectors any of the set of candidate intent vectors that are semantically similar to the input intent vector.