FIELD: recognition technology.
SUBSTANCE: invention relates to the field of computer technology for recognizing handwritten text. The expected result is achieved by a text recognition method that includes: obtaining an image with a text string; segmentation of the image into two or more image fragments; for each of two or more image fragments, determining the first hypothesis for segmenting an image fragment into the first set of grapheme images; determining the first fragmentation confidence score for the first hypothesis; determining the second hypothesis for segmenting an image fragment into a second set of grapheme images; determining the second fragmentation confidence score for the second hypothesis; determining that the first fragmentation confidence score is higher than the second fragmentation confidence score; and converting the first set of grapheme images defined by the first hypothesis into a set of characters; and collecting the set of characters of each image fragment to obtain a string of text.
EFFECT: improved character recognition.
20 cl, 10 dwg
| Title | Year | Author | Number |
|---|---|---|---|
| NEURAL NETWORK TRAINING BY MEANS OF SPECIALIZED LOSS FUNCTIONS | 2018 |
|
RU2707147C1 |
| OPTICAL CHARACTER RECOGNITION USING SPECIALIZED CONFIDENCE FUNCTIONS, IMPLEMENTED ON THE BASIS OF NEURAL NETWORKS | 2018 |
|
RU2703270C1 |
| DIFFERENTIAL CLASSIFICATION WITH MULTIPLE NEURAL NETWORKS | 2017 |
|
RU2652461C1 |
| CHARACTER RECOGNITION USING A HIERARCHICAL CLASSIFICATION | 2018 |
|
RU2693916C1 |
| IDENTIFICATION OF WRITING SYSTEMS USED IN DOCUMENTS | 2021 |
|
RU2792743C1 |
| IDENTIFICATION OF FIELDS ON AN IMAGE USING ARTIFICIAL INTELLIGENCE | 2018 |
|
RU2695489C1 |
| METHOD AND SYSTEM FOR EXTRACTING DATA FROM IMAGES OF SEMISTRUCTURED DOCUMENTS | 2015 |
|
RU2613846C2 |
| OPTICAL CHARACTER RECOGNITION BY MEANS OF COMBINATION OF NEURAL NETWORK MODELS | 2020 |
|
RU2768211C1 |
| DETECTING TEXT FIELDS USING NEURAL NETWORKS | 2018 |
|
RU2699687C1 |
| METHOD OF DETECTING NECESSITY OF STANDARD LEARNING FOR VERIFICATION OF RECOGNIZED TEXT | 2014 |
|
RU2641225C2 |
