About Yandex Vision OCR

OCR stands for optical character recognition. Yandex Vision OCR is a computer vision service that enables image text and PDF recognition.

Vision OCR provides its features through API. You can integrate Vision OCR features into your app written in any language or send requests using cURL or similar utilities. Vision OCR provides API in REST and gRPC notations. You can generate your own API for your programming language using this API specification.

Vision OCR operating modesVision OCR operating modes

Vision OCR can process image recognition requests both synchronously and asynchronously.

• In synchronous mode, Vision OCR will process your request once it gets it and will provide you with the result in the response. This mode is good for apps that need to communicate with the user. However, you cannot use Vision OCR synchronous mode to process large pieces of information.
• In asynchronous mode, Vision OCR will get your request and immediately return the operation ID you can use to get the recognition result. Recognizing text in asynchronous mode takes more time but allows processing large volumes of information in a single request. Use asynchronous mode if you do not need an urgent response.

Recognition modelsRecognition models

Vision OCR provides various models to recognize different types of image and PDF text. In particular, there are models for normal text, multi-column text, tables, handwritten text, or common documents, such as passport or license plate number. With a more suitable model, you get better recognition result. To specify the model you need, use the model field in your request.

See below for the list of available recognition models:

• page (default): Suitable for images with any number of text lines within a single column.
• page-column-sort: Use it to recognize multi-column text.
• handwritten: Use it to recognize a combination of typed and handwritten text in English or Russian.
• table: Use it to recognize tables in English or Russian.

Models for recognizing common documents:

• passport: Passport data page.
• driver-license-front: Driver license, front side.
• driver-license-back: Driver license, back side.
• vehicle-registration-front: Vehicle registration certificate, front side.
• vehicle-registration-back: Vehicle registration certificate, back side.
• license-plates: All license plate numbers in the image.

Language model detectionLanguage model detection

For text recognition, Vision OCR uses language models trained based on specific languages. Vision OCR selects a suitable model automatically from the list you provide in your request.

Only a single model is used each time you recognize a text. For example, if an image contains text in Chinese and Japanese, only one language will be recognized. To recognize both, send another request specifying the other language.

Image requirementsImage requirements

An image in a request must meet the following requirements:

• The supported file formats are JPEG, PNG, and PDF. Specify the MIME type of the file in the mime_type property. The default value is image.
• The maximum file size is 10 MB.
• The image size should not exceed 20 MP (height × width).

Response with recognition resultsResponse with recognition results

The service highlights the text characters found in the image and groups them by level: words are grouped into lines, lines into blocks, and blocks into pages.

As a result, Yandex Vision OCR returns an object with the following properties:

• For pages[]: Page size.
• For blocks[]: Position of the text on the page.
• For lines[]: Position of lines.
• For words[]: Position of words, text, and language used for recognition.

To show the position of the text, Yandex Vision OCR returns the coordinates of the rectangle that frames the text. Coordinates are the number of pixels from the top-left corner of the image.

The coordinates of a rectangle are calculated from the top-left corner and specified counterclockwise:

1←4
↓ ↑
2→3

Here is an example of a recognized image with coordinates:

{
  "result": {
    "text_annotation": {
      "width": "1920",
      "height": "1280",
      "blocks": [{
        "bounding_box": {
          "vertices": [{
            "x": "460",
            "y": "777"
          }, {
            "x": "460",
            "y": "906"
          }, {
            "x": "810",
            "y": "906"
          }, {
            "x": "810",
            "y": "777"
          }]
        },
        "lines": [{
          "bounding_box": {
            "vertices": [{
              "x": "460",
              "y": "777"
            }, {
              "x": "460",
              "y": "820"
            }, {
              "x": "802",
              "y": "820"
            }, {
              "x": "802",
              "y": "777"
            }]
          },
          "alternatives": [{
            "text": "PENGUINS",
            "words": [{
              "bounding_box": {
                "vertices": [{
                  "x": "460",
                  "y": "768"
                }, {
                  "x": "460",
                  "y": "830"
                }, {
                  "x": "802",
                  "y": "830"
                }, {
                  "x": "802",
                  "y": "768"
                }]
              },
              "text": "PENGUINS",
              "entity_index": "-1"
            }]
          }]
        }, {
          "bounding_box": {
            "vertices": [{
              "x": "489",
              "y": "861"
            }, {
              "x": "489",
              "y": "906"
            }, {
              "x": "810",
              "y": "906"
            }, {
              "x": "810",
              "y": "861"
            }]
          },
          "alternatives": [{
            "text": "CROSSING",
            "words": [{
              "bounding_box": {
                "vertices": [{
                  "x": "489",
                  "y": "852"
                }, {
                  "x": "489",
                  "y": "916"
                }, {
                  "x": "810",
                  "y": "916"
                }, {
                  "x": "810",
                  "y": "852"
                }]
              },
              "text": "CROSSING",
              "entity_index": "-1"
            }]
          }]
        }],
        "languages": [{
          "language_code": "en"
        }]
      }, {
        "bounding_box": {
          "vertices": [{
            "x": "547",
            "y": "989"
          }, {
            "x": "547",
            "y": "1046"
          }, {
            "x": "748",
            "y": "1046"
          }, {
            "x": "748",
            "y": "989"
          }]
        },
        "lines": [{
          "bounding_box": {
            "vertices": [{
              "x": "547",
              "y": "989"
            }, {
              "x": "547",
              "y": "1046"
            }, {
              "x": "748",
              "y": "1046"
            }, {
              "x": "748",
              "y": "989"
            }]
          },
          "alternatives": [{
            "text": "SLOW",
            "words": [{
              "bounding_box": {
                "vertices": [{
                  "x": "547",
                  "y": "983"
                }, {
                  "x": "547",
                  "y": "1054"
                }, {
                  "x": "748",
                  "y": "1054"
                }, {
                  "x": "748",
                  "y": "983"
                }]
              },
              "text": "SLOW",
              "entity_index": "-1"
            }]
          }]
        }],
        "languages": [{
          "language_code": "en"
        }]
      }],
      "entities": []
    },
    "page": "0"
  }
}

Response formatResponse format

Yandex Vision OCR provides recognition results in JSON Lines format, where each line of a JSON file corresponds to one recognized page or image.

Errors in determining coordinatesErrors in determining coordinates

Coordinates returned by the service may in some cases mismatch the text displayed in the user's image processor. This is due to incorrect handling of exif metadata by the user's image processor.

During recognition, Yandex Vision OCR considers data about image rotation set by the Orientation attribute in the exif section. Some tools used for viewing images may ignore the rotation values set in exif. This causes a mismatch between the obtained results and the displayed image.

To fix this error, do one of the following:

• Change the image processor settings so that the rotation angle specified in the exif section is considered while viewing images.
• Remove the Orientation attribute from the image exif section or set it to 0 when providing the image to Yandex Vision OCR.

What's nextWhat's next

• See the list of supported languages
• Learn about known issues in the current version
• Try recognizing text in an image