Search International and National Patent Collections

1. (WO2017031716) METHOD FOR ANALYZING AND RECOGNIZING HANDWRITTEN MATHEMATICAL FORMULA STRUCTURE IN NATURAL SCENE IMAGE

Pub. No.:    WO/2017/031716    International Application No.:    PCT/CN2015/088113
Publication Date: Fri Mar 03 00:59:59 CET 2017 International Filing Date: Thu Aug 27 01:59:59 CEST 2015
IPC: G06K 9/32
G06K 9/62
G06N 3/02
Applicants: BEIJING LEJENT TECHNOLOGY CO., LTD
北京云江科技有限公司
Inventors: CHEN, Li jiang
陈李江
LIU, Ning
刘宁
LIU, Hui
刘辉
Title: METHOD FOR ANALYZING AND RECOGNIZING HANDWRITTEN MATHEMATICAL FORMULA STRUCTURE IN NATURAL SCENE IMAGE
Abstract:
A method for analyzing and recognizing handwritten mathematical formula structure in natural scene image comprises: S1, converting a gray level matrix of a natural scene image into a local contrast matrix, and performing binary division on the partial contrast matrix by using an otsu method, to obtain a binary matrix; S2, performing connected component analysis on the binary matrix in step S1, to eliminate non-character connected components and obtain character connected components; S3, detecting formula special-structure elements in the character connected components in S2 by using a correlation coefficient method, and separately labeling all the detected special-structure elements; S4, dividing the binary matrix in S1 by using a horizontal projection method; S5: recognizing each character connected component by using a convolutional neural network; and S6, defining an output sequence, and outputting recognition results, in a latex typesetting format, according to the corresponding sequence. The method effectively solves the expression problem of elementary mathematical formulas in OCR recognition.