Introduction

git clone git@github.com:Mathpix/api-examples.git
cd api-examples/images

The MathpixOCR API is a JSON API for extracting text from images and digital ink inputs. Unlike other OCR API's, MathpixOCR has 1rst class support for scientific notation, as used in chemistry, math, physics, computer science, economics, and other STEM subjects.

If you have any questions or problems, please send us an email at support@mathpix.com.

Authorization

The request headers must be set as follows:

{
  "content-type": "application/json",
  "app_id": "YOUR_APP_ID",
  "app_key": "YOUR_APP_KEY"
}

MathpixOCR uses API keys to allow access to the API. You can find your API keys on your account dashboard at https://accounts.mathpix.com/ocr-api.

MathpixOCR expects for the API key to be included in all API requests to the server via HTTP Basic Auth. Expected set of HTTP headers is shown on the right.

Formatting

All text outputs from the API conform to the Mathpix Markdown spec. Currently, our OCR will emit block mode math \[ ... \] and inline math $ ... $.

You can use mathpix-markdown in your app by using our open source library mathpix-markdown-it which is available via NPM.

The mathpix-markdown-it library is also capable of generating HTML and structured data containing alternative data representations such as MathML, Asciimath, for math, and TSV (tab separate values) for tables.

If you do not wish to use mathpix-markdown-it to render text / latex, you may use MathJax, Katex, and PDFLaTeX.

Process image (v3/text)

Request image:

Request JSON:

{
    "src": "https://mathpix.com/examples/limit.jpg",
    "formats": ["text", "data", "html"],
    "data_options": {
        "include_asciimath": true,
        "include_latex": true
    }
}

Response:

{
    "confidence": 1,
    "confidence_rate": 1,
    "is_printed": false, 
    "is_handwritten": true,
    "text": "\\( \\lim _{x \\rightarrow 3}\\left(\\frac{x^{2}+9}{x-3}\\right) \\)",
    "html": "<div><span class=\"math-inline \">\n<asciimath style=\"display: none;\">lim_(x rarr3)((x^(2)+9)/(x-3))</asciimath><latex style=\"display: none\">\\lim _{x \\rightarrow 3}\\left(\\frac{x^{2}+9}{x-3}\\right)</latex></span></div>\n",
    "data": [
        {
            "type": "asciimath",
            "value": "lim_(x rarr3)((x^(2)+9)/(x-3))"
        },
        {
            "type": "latex",
            "value": "\\lim _{x \\rightarrow 3}\\left(\\frac{x^{2}+9}{x-3}\\right)"
        }
    ]
}

Mathpix supports image recognition for jpg and png images. Images are sent via URL, or are sent via base64 encoded images.

The v3/text endpoint extracts text, and optionally derived data / HTML, from images.

The text outputs follow mathpix-markdown conventions, including math mode Latex inside inline delimiters $ ... $ and block mode delimiters \[ .... \]. Lines are separated with \n newline characters. In some cases (eg multiple choice equations) we will try flatten horizontally aligned content into different lines in order to keep the markup simple.

We also provide structured data outputs via the data and html output options. The data output returns a list of extracted formats (such as tsv for tables, or asciimath for equations). The html output provides annotated HTML and can be parsed via HTML / XML parsers.

Request parameters

Send an image:

{
    "src": "data:image/jpeg;base64,...",
    "formats": ["text", "data", "html"],
    "data_options": {
        "include_asciimath": true,
        "include_latex": true
    }
}

#!/usr/bin/env python
import sys
import base64
import requests
import json

# put desired file path here
file_path = 'limit.jpg'
image_uri = "data:image/jpg;base64," + base64.b64encode(open(file_path, "rb").read()).decode()
r = requests.post("https://api.mathpix.com/v3/text",
    data=json.dumps({'src': image_uri}),
    headers={"app_id": "YOUR_APP_ID", "app_key": "YOUR_APP_KEY",
             "Content-type": "application/json"})
print(json.dumps(json.loads(r.text), indent=4, sort_keys=True))

curl -X POST https://api.mathpix.com/v3/text \
-H 'app_id: YOUR_APP_ID' \
-H 'app_key: YOUR_APP_KEY' \
-H 'Content-Type: application/json' \
--data '{ "src": "data:image/jpeg;base64,'$(base64 -i limit.jpg)'" }'

POST https://api.mathpix.com/v3/text

Parameter	Type	Description
`src`	string	Image data, or public URL where image is located
`metadata` (optional)	object	Key value object
`formats` (optional)	[string]	List of formats, one of `text`, `data`, `html`, `latex_styled`, see Format Descriptions
`data_options` (optional)	object	See DataOptions section, specifies outputs for `data` and `html` return fields
`include_detected_alphabets` (optional)	bool	Return detected alphabets
`alphabets_allowed` (optional)	object	See AlphabetsAllowed section, use this to specify which alphabets you don't want in the output
`confidence_threshold` (optional)	number in [0,1]	Specifies threshold for triggering confidence errors
`confidence_rate_threshold` (optional)	number in [0,1]	Specifies threshold for triggering confidence errors, default 0.75 (symbol level threshold)
`include_line_data` (optional)	bool	Specifies whether to return information per line segmented, see LineData object section for details
`auto_rotate_confidence_threshold` (optional)	number in [0,1]	Specifies threshold for auto rotating image to correct orientation; by default it is set to 0.99, can be disabled with a value of 1 (see Auto rotation section for details)
`rm_spaces` (optional)	bool	Determines whether extra white space is removed from equations in `latex_styled` and `text` formats. Default is `true`.
`rm_fonts` (optional)	bool	Determines whether font commands such as \mathbf and \mathrm are removed from equations in `latex_styled` and `text` formats. Default is `false`.
`numbers_default_to_math` (optional)	bool	Specifies whether numbers are always math, e.g., `Answer: $ 17 $` instead of `Answer: 17`. Default is `false`.

Format descriptions

Mathpix OCR returns the following string formats from images:

Format	Description
`text`	Mathpix markdown formatted text
`html`	HTML rendered from `text` via mathpix-markdown-it
`data`	Data extracted from `html` as specified in the `data_options` request parameter
`latex_styled`	Styled Latex, returned only in cases that the whole image can be reduced to a single equation

DataOptions object

Data options are used to parse relevant information from the image output. These outputs are all computed directly from the html format described above.

Key	Type	Description
`include_svg` (optional)	bool	include math SVG in `html` and `data` formats
`include_table_html` (optional)	bool	include HTML for `html` and `data` outputs (tables only)
`include_latex` (optional)	bool	include math mode latex in `data` and `html`
`include_tsv` (optional)	bool	include tab separated values (TSV) in `data` and `html` outputs (tables only)
`include_asciimath` (optional)	bool	include asciimath in `data` and `html` outputs
`include_mathml` (optional)	bool	include mathml in `data` and `html` outputs

Result object

Get an API response:

{
    "confidence": 0.9982182085336344,
    "confidence_rate": 0.9982182085336344,
    "is_printed": false, 
    "is_handwritten": true,
    "data": [
        {
            "type": "asciimath",
            "value": "lim_(x rarr3)((x^(2)+9)/(x-3))"
        },
        {
            "type": "latex",
            "value": "\\lim _{x \\rightarrow 3}\\left(\\frac{x^{2}+9}{x-3}\\right)"
        }
    ],
    "html": "<div><span  class=\"math-inline \" >\n<asciimath style=\"display: none;\">lim_(x rarr3)((x^(2)+9)/(x-3))</asciimath><latex style=\"display: none\">\\lim _{x \\rightarrow 3}\\left(\\frac{x^{2}+9}{x-3}\\right)</latex></span></div>\n",
    "text": "\\( \\lim _{x \\rightarrow 3}\\left(\\frac{x^{2}+9}{x-3}\\right) \\)"
}

Field	Type	Description
`request_id`	string	Request ID, for debugging purposes
`text` (optional)	string	Recognized `text` format, if such is found
`latex_styled` (optional)	string	Math Latex string of math equation, if the image is of a single equation
`confidence` (optional)	number in [0,1]	Estimated probability 100% correct
`confidence_rate` (optional)	number in [0,1]	Estimated confidence of input quality
`line_data` (optional)	[object]	List of LineData objects
`data` (optional)	[object]	List of Data objects
`html` (optional)	string	Annotated HTML output
`detected_alphabets` (optional)	[object]	DetectedAlphabet object
`is_printed` (optional)	bool	Specifies if printed content was detected in an image
`is_handwritten` (optional)	bool	Specifies if handwritten content was detected in an image
`auto_rotate_confidence` (optional)	number in [0,1]	Estimated probability that image needs to be rotated, see Auto rotation
`auto_rotate_degrees` (optional)	number in {0, 90, -90, 180}	Estimated angle of rotation in degrees to put image in correct orientation, see Auto rotation
`error` (optional)	string	US locale error message
`error_info` (optional)	object	Error info object

Data object

Data objects allow extracting relevant data from an OCR result.

Field	Type	Description
`type`	string	one of `asciimath`, `mathml`, `latex`, `html`, `svg`, `tsv`
`value`	string	value corresponding to `type`

DetectedAlphabet object

The detected_alphabets object in a result contains a field that is true of false for each known alphabet. The field is true if any characters from the alphabet are recognized in the image, regardless of whether any of the result fields contain the characters.

Field	Type	Description
`en`	bool	English
`hi`	bool	Hindi Devenagari
`zh`	bool	Chinese
`ja`	bool	Kana Hiragana or Katakana
`ko`	bool	Hangul Jamo
`ru`	bool	Russian
`th`	bool	Thai

AlphabetsAllowed object

{
    "src": "data:image/jpeg;base64,...",
    "formats": ["text"],
    "alphabets_allowed": {
          "hi": false,
          "zh": false,
          "ja": false, 
          "ko": false,
          "ru": false,
          "th": false
        }
}

There are some cases where it is not easy to infer the correct alphabet for a single letter, because there are different letters from different alphabets that look alike. To illustrate, one example is conflict between Latin B and Cyrillic В (that is Latin V). While being displayed almost the same, they essentially have different Unicode encodings. The option alphabets_allowed can be used to specify map from string to boolean values which can be used to prevent symbols from unwanted alphabet appearing in the result. Map keys that are valid correspond to the values in Field column of the table specified in Detected alphabet object section (e.g. hi or ru). By default all alphabets are allowed in the output, to disable alphabet specify "alphabets_allowed": {"alphabet_key": false}.

Specifying "alphabets_allowed": {"alphabet_key": true} has the same effect as not specifying that alphabet inside alphabets_allowed map.

LineData object

Example request:

{
    "src": "https://mathpix.com/examples/text_with_diagram.png",
    "formats": ["text"],
    "include_line_data": true
}

JSON response:

{
    "confidence": 0.651358435330524,
    "confidence_rate": 0.651358435330524,
    "text": "Equivalent resistance between points \\( \\mathrm{A} \\& \\mathrm{B} \\) in the adjacent circuit is",
    "line_data": [
        {
            "type": "text",
            "cnt": [
                [
                    859,
                    81
                ],
                [
                    739,
                    91
                ],
                [
                    626,
                    91
                ],
                [
                    -2,
                    66
                ],
                [
                    0,
                    34
                ],
                [
                    739,
                    52
                ],
                [
                    859,
                    63
                ]
            ],
            "included": true,
            "text": "Equivalent resistance between points \\( \\mathrm{A} \\& \\mathrm{B} \\) in the adjacent circuit is",
            "after_hyphen": false,
            "confidence": 0.651358435330524,
            "confidence_rate": 0.9948483133235457
        },
        {
            "type": "diagram",
            "cnt": [
                [
                    654,
                    244
                ],
                [
                    651,
                    683
                ],
                [
                    7,
                    678
                ],
                [
                    11,
                    238
                ]
            ],
            "included": false,
            "error_id": "image_not_supported"
        }
    ]
}

Field	Type	Description
`type`	string	One of `text`, `math`, `table`, `diagram`, `equation_number`
`subtype` (optional)	string	Either not set, or `chemistry`, or `triangle` (more diagram subtypes coming soon)
`cnt`	[[x,y]]	Countour for line expressed as list of (x,y) pixel coordinate pairs
`included`	bool	Whether this line is included in the top level OCR result
`error_id` (optional)	string	Error ID, reason why the line is not included in final result
`text` (optional)	string	Text (Mathpix Markdown) for line
`confidence` (optional)	number in [0,1]	Estimated probability 100% correct
`confidence_rate` (optional)	number in [0,1]	Estimated confidence of input quality
`after_hyphen` (optional)	bool	specifies if the current line occurs after the text line which ended with hyphen
`html` (optional)	string	Annotated HTML output for the line
`data` (optional)	[Data]	List of Data object's

The v3/text endpoint allows customers to request line by line data by adding a include_line_data request parameter to the request. When this parameter is true, the response object then includes a line_data field which is a list of LineData objects containing information about all texual line elements detected in the image. Simply concatenating information from the response's line_data is enough to recreate the top level text, html, and data fields included in the response JSON.

Some lines are not supported by the OCR engine, like diagrams, and are therefore simply skipped by the OCR engine. Some lines contain content that is most likely extraneous, like equation numbers. Additionally, sometimes the OCR engine simply cannot recognize the line with proper confidence. In all those cases included field is set to false, as that line is certainly not part of the final result.

The following error_id's can occur here:

image_not_supported - OCR engine doesn't accept this type of line
image_max_size - line is larger than maximal size which OCR engine supports
math_confidence - OCR engine failed to confidently recognize the content of the line
image_no_content - line has strange spatial dimensions, e.g. height of the line is zero; this error is very unlikely to happen

Auto rotation

Sometimes images that are received are in wrong orientation,one example of such image would look like this:

The goal of automatic rotation is to pick correct orientation for received images before any processing is done. The result of auto rotation looks like:

One can control how confident our algorithm needs to be to perform auto rotation with request parameter auto_rotate_confidence_threshold, a number in [0,1]. By default the value 0.99 is used, which essentially means that algorithm will auto rotate image if it is 99% confident in that decision. Auto rotation can be disabled by specifying "auto_rotate_confidence_threshold": 1 as a part of request being sent.

Response object includes two related fields:

auto_rotate_confidence - confidence of the algorithm in decision that image needs to be rotated, number in [0,1] which should be ~0 if image is in correct orientation and ~1 if image should be rotated; monitoring this value on custom workloads can be helpful in determining proper auto_rotate_confidence_threshold to be used with such workloads
auto_rotate_degrees - angle in degrees specifying how much to rotate original image to get image in proper orientation, number in {0, 90, -90, 180}, value 0 means image is already in correct orientation; this information can be used to put image in right orientation for other parts of custom processing pipelines

Process strokes (v3/strokes)

Mathpix supports handwriting recognition for strokes coordinates.

The v3/strokes endpoint is in beta but provides a service able to transform handwritten strokes into its transcript of text and math.

This endpoint is very convenient for users that were generating images of handwritten math and text and then using the service v3/text, since with v3/strokes no image generation is required, the request payload is smaller and therefore it results in faster response time.

The LaTeX of the recognized handwriting is returned inside inline delimiters $ ... $ and block mode delimiters \[ .... \]. Lines are separated with \n newline characters. In some cases (e.g. multiple choice equations) we will try to flatten horizontally aligned content into different lines in order to keep the markup simple.

Request parameters

Send some strokes:

{
  "strokes": {"strokes": {
    "x": [[131,131,130,130,131,133,136,146,151,158,161,162,162,162,162,159,155,147,142,137,136,138,143,160,171,190,197,202,202,202,201,194,189,177,170,158,153,150,148],[231,231,233,235,239,248,252,260,264,273,277,280,282,283],[273,272,271,270,267,262,257,249,243,240,237,235,234,234,233,233],[296,296,297,299,300,301,301,302,303,304,305,306,306,305,304,298,294,286,283,281,281,282,284,284,285,287,290,293,294,299,301,308,309,314,315,316]],
    "y": [[213,213,212,211,210,208,207,206,206,209,212,217,220,227,230,234,236,238,239,239,239,239,239,239,241,247,252,259,261,264,266,269,270,271,271,271,270,269,268],[231,231,232,235,238,246,249,257,261,267,270,272,273,274],[230,230,230,231,234,240,246,258,268,273,277,281,281,283,283,284],[192,192,191,189,188,187,187,187,188,188,190,193,195,198,200,205,208,213,215,215,215,214,214,214,214,216,218,220,221,223,223,223,223,221,221,220]]
  }}
}

#!/usr/bin/env python
import sys
import base64
import requests
import json

# put input strokes here
strokes_string = '{"strokes": {\
    "x": [[131,131,130,130,131,133,136,146,151,158,161,162,162,162,162,159,155,147,142,137,136,138,143,160,171,190,197,202,202,202,201,194,189,177,170,158,153,150,148],[231,231,233,235,239,248,252,260,264,273,277,280,282,283],[273,272,271,270,267,262,257,249,243,240,237,235,234,234,233,233],[296,296,297,299,300,301,301,302,303,304,305,306,306,305,304,298,294,286,283,281,281,282,284,284,285,287,290,293,294,299,301,308,309,314,315,316]],\
    "y": [[213,213,212,211,210,208,207,206,206,209,212,217,220,227,230,234,236,238,239,239,239,239,239,239,241,247,252,259,261,264,266,269,270,271,271,271,270,269,268],[231,231,232,235,238,246,249,257,261,267,270,272,273,274],[230,230,230,231,234,240,246,258,268,273,277,281,281,283,283,284],[192,192,191,189,188,187,187,187,188,188,190,193,195,198,200,205,208,213,215,215,215,214,214,214,214,216,218,220,221,223,223,223,223,221,221,220]]\
  }}'
strokes = json.loads(strokes_string)
r = requests.post("https://api.mathpix.com/v3/strokes",
    data=json.dumps({'strokes': strokes}),
    headers={"app_id": "YOUR_APP_ID", "app_key": "YOUR_APP_KEY",
             "Content-type": "application/json"})
print(json.dumps(json.loads(r.text), indent=4, sort_keys=True))

curl -X POST https://api.mathpix.com/v3/strokes \
-H 'app_id: YOUR_APP_ID' \
-H 'app_key: YOUR_APP_KEY' \
-H 'Content-Type: application/json' \
--data '{ "strokes": {"strokes": {
  "x": [[131,131,130,130,131,133,136,146,151,158,161,162,162,162,162,159,155,147,142,137,136,138,143,160,171,190,197,202,202,202,201,194,189,177,170,158,153,150,148],[231,231,233,235,239,248,252,260,264,273,277,280,282,283],[273,272,271,270,267,262,257,249,243,240,237,235,234,234,233,233],[296,296,297,299,300,301,301,302,303,304,305,306,306,305,304,298,294,286,283,281,281,282,284,284,285,287,290,293,294,299,301,308,309,314,315,316]],
  "y": [[213,213,212,211,210,208,207,206,206,209,212,217,220,227,230,234,236,238,239,239,239,239,239,239,241,247,252,259,261,264,266,269,270,271,271,271,270,269,268],[231,231,232,235,238,246,249,257,261,267,270,272,273,274],[230,230,230,231,234,240,246,258,268,273,277,281,281,283,283,284],[192,192,191,189,188,187,187,187,188,188,190,193,195,198,200,205,208,213,215,215,215,214,214,214,214,216,218,220,221,223,223,223,223,221,221,220]]
  }}}'

POST https://api.mathpix.com/v3/strokes

Parameter	Type	Description
`strokes`	JSON	Strokes in JSON with appropriate format.
`metadata` (optional)	object	Key value object
`formats` (optional)	[string]	List of formats, one of `text`, `data`, `html`
`data_options` (optional)	object	see "Data options" section above, specifies outputs for `data` and `html` return fields

Result object

Get an API response:

{
    "text": "\\( 3 x^{2} \\)",
    "confidence": 0.9999953508431929,
    "confidence_rate": 0.9999953508431929
}

Field	Type	Description
`text` (optional)	string	Recognized `text` format, if such is found
`confidence` (optional)	number in [0,1]	Estimated probability 100% correct
`confidence_rate` (optional)	number in [0,1]	Estimated confidence of input quality
`data` (optional)	[object]	List of data objects (see "Data object" section above)
`html` (optional)	string	Annotated HTML output

Process batch (v3/batch)

A batch request is made with JSON that looks like:

{  
   "urls":{  
      "inverted":"https://raw.githubusercontent.com/Mathpix/api-examples/master/images/inverted.jpg",
      "algebra":"https://raw.githubusercontent.com/Mathpix/api-examples/master/images/algebra.jpg"
   },
   "formats": ["latex_simplified"]
}

curl -X POST https://api.mathpix.com/v3/batch \
-H "app_id: YOUR_APP_ID" \
-H "app_key: YOUR_APP_KEY" \
-H "Content-Type: application/json" \
--data '{ "urls": {"inverted": "https://raw.githubusercontent.com/Mathpix/api-examples/master/images/inverted.jpg", "algebra": "https://raw.githubusercontent.com/Mathpix/api-examples/master/images/algebra.jpg"},"formats":["latex_simplified"] }'


import requests
import json

base_url = 'https://raw.githubusercontent.com/Mathpix/api-examples/master/images/'

data = {
    'urls': {
        'algebra': base_url + 'algebra.jpg',
        'inverted': base_url + 'inverted.jpg'
    },
    'formats': ['latex_simplified']
}

r = requests.post(
    "https://api.mathpix.com/v3/batch", data=json.dumps(data),
    headers={
        'app_id': 'YOUR_APP_ID',
        'app_key': 'YOUR_APP_KEY',
        'content-type': 'application/json'
    },
    timeout=30
)
reply = json.loads(r.text)
assert reply.has_key('batch_id')

The response to the batch is a positive integer value for batch_id.

{
  "batch_id": "17"
}

The response to GET /v3/batch/17 is below when the batch has completed. Before completion the "results" field may be empty or contain only one of the two results.

{
  "keys": ["algebra", "inverted"],
  "results": {
    "algebra": {
      "detection_list": [], 
      "detection_map": {
        "contains_chart": 0, 
        "contains_diagram": 0, 
        "contains_geometry": 0, 
        "contains_graph": 0, 
        "contains_table": 0, 
        "is_inverted": 0, 
        "is_not_math": 0, 
        "is_printed": 0
      }, 
      "latex_simplified": "12 + 5 x - 8 = 12 x - 10", 
      "latex_confidence": 0.99640350138238, 
      "position": {
        "height": 208, 
        "top_left_x": 0, 
        "top_left_y": 0, 
        "width": 1380
      }
    }, 
    "inverted": {
      "detection_list": [
        "is_inverted", 
        "is_printed"
      ], 
      "detection_map": {
        "contains_chart": 0, 
        "contains_diagram": 0, 
        "contains_geometry": 0, 
        "contains_graph": 0, 
        "contains_table": 0, 
        "is_inverted": 1, 
        "is_not_math": 0, 
        "is_printed": 1
      }, 
      "latex_simplified": "x ^ { 2 } + y ^ { 2 } = 9", 
      "latex_confidence": 0.99982263230866, 
      "position": {
        "height": 170, 
        "top_left_x": 48, 
        "top_left_y": 85, 
        "width": 544
      }
    }
  }
}

The Mathpix API supports processing multiple images in a single POST request to a different endpoint: /v3/batch. The request body may contain all the /v3/latex parameters except src and must contain a urls parameter. The request may contain an additonal callback parameter to receive results after all the images in the batch have been processed.

Parameter	Type	Description
urls	object	key-value for each image in the batch where the value may be a string url or an object containing a url and image-specific request arguments such as `region` and `formats`.
callback (optional)	object	description of where to send the batch results
callback.post	string	url to post results
callback.reply (optional)	object	data to send in reply to batch POST
callback.body (optional)	object	data to send with results to callback.post
callback.headers (optional)	object	headers to use when posting results

The response contains only a unique batch_id value. Even if the request includes a callback, there is no guarantee the callback will run successfuly (because of a transient network failure, for example). The preferred approach is to wait an appropriate length of time (about one second for every five images in the batch) and then do a GET on /v3/batch/:id where :id is the batch_id value. The GET request must contain the same app_id and app_key headers as the POST to /v3/batch.

The GET response has the following fields:

Field	Type	Description
keys	string[]	all the url keys present in the originating batch request
results	object	an OCR result for each key that has been processed

Legacy endpoint (v3/latex)

This is a legacy endpoint. We recommend using v3/text or v3/strokes instead.

Mathpix supports image recognition for jpg and png images. Images are encoded by base64 and sent inside JSON requests.

Request parameters

You can request multiple formats for a single image:

{
  "src": "data:image/jpeg;base64,...",
  "ocr": ["math", "text"], 
  "skip_recrop": true,
  "formats": [
    "text",
    "latex_simplified",
    "latex_styled",
    "mathml",
    "asciimath",
    "latex_list"
   ]
}

#!/usr/bin/env python
import sys
import base64
import requests
import json

# put desired file path here
file_path = 'limit.jpg'
image_uri = "data:image/jpg;base64," + base64.b64encode(open(file_path, "rb").read()).decode()
r = requests.post("https://api.mathpix.com/v3/latex",
    data=json.dumps({'src': image_uri, 'formats': ['latex_normal']}),
    headers={"app_id": "YOUR_APP_ID", "app_key": "YOUR_APP_KEY",
             "Content-type": "application/json"})
print(json.dumps(json.loads(r.text), indent=4, sort_keys=True))

curl -X POST https://api.mathpix.com/v3/latex \
-H 'app_id: YOUR_APP_ID' \
-H 'app_key: YOUR_APP_KEY' \
-H 'Content-Type: application/json' \
--data '{ "src": "data:image/jpeg;base64,'$(base64 -i limit.jpg)'", "formats": ["latex_normal"] }'

POST https://api.mathpix.com/v3/latex

Parameter	Type	Description
src	string	Image data, or public URL where image is located
formats	string[]	String postprocessing formats (see Formatting section)
ocr (optional)	string[]	Process only math ["math"] or both math and text ["math", "text"]
format_options (optional)	object	Options for specific formats (see Formatting section)
skip_recrop (optional)	bool	Force algorithm to consider whole image
confidence_threshold (optional)	number in [0,1]	Set threshold for triggering confidence errors
beam_size (optional)	number in [1,5]	Number of results to consider during recognition
n_best (optional)	number in [1,beam_size]	Number of highest-confidence results to return
region (optional)	object	Specify the image area with the pixel coordinates `top_left_x`, `top_left_y`, `width`, and `height`
callback (optional)	object	Callback request object
metadata (optional)	object	Key value object
include_detected_alphabets (optional)	bool	Return detected alphabets
auto_rotate_confidence_threshold (optional)	number in [0,1]	Specifies threshold for auto rotating image to correct orientation; by default it is set to 0.99, can be disabled with a value of 1 (see Auto rotation section for details)

Formatting

The following formats can be used in the request:

Format	Description
text	text mode output, with math inside delimiters, eg. `test $x^2$`, inline math by default
text_display	same as `text`, except uses block mode math instead of inline mode when in doubt
latex_normal	direct LaTeX representation of the input
latex_styled	modified output to improve the visual appearance such as adding '\left' and '\right' around parenthesized expressions that contain tall expressions like subscript or superscript
latex_simplified	modified output for symbolic processing such as shortening operator names, replacing long division with a fraction, and converting a column of operands into a single formula
latex_list	output split into a list of simplified strings to help process multiple equations
mathml	the MathML for the recognized math
asciimath	the AsciiMath for the recognized math
wolfram	a string compatible with the Wolfram Alpha engine

Format options

To return a more compact latex_styled result, one could send the following request:

{
  "src":"data:image/jpeg;base64,...",
  "ocr": ["math", "text"], 
  "skip_recrop": true,
  "formats": [
    "text",
    "latex_simplified",
    "latex_styled",
    "mathml",
    "asciimath",
    "latex_list"
   ],
   "format_options": {
     "latex_styled": {"transforms": ["rm_spaces"]}
   }
}

The result for "latex_styled" would now be
"\\lim_{x \\rightarrow 3}\\left(\\frac{x^{2}+9}{x-3}\\right)"
instead of
"\\lim _ { x \\rightarrow 3 } \\left( \\frac { x ^ { 2 } + 9 } { x - 3 } \\right)"

The optional format_options request parameter allows a request to customize the LaTeX result formats using an object with a format as the property name and the options for that format as the value. The options value may specify the following properties:

Option	Type	Description
`transforms`	string[]	array of transformation names
`math_delims`	[string, string]	[begin, end] delimiters for math mode, for example `$` and `$`
`displaymath_delims`	[string, string]	[begin, end] delimiters for displaymath mode, for example `\[` and `\]`

The currently-supported transforms are:

Transform	Description
rm_spaces	omit spaces around LaTeX groups and other places where spaces are superfluous
rm_newlines	uses spaces instead of newlines between text lines in paragraphs
rm_fonts	omit mathbb, mathbf, mathcal, and mathrm commands
rm_style_syms	replace styled commands with unstyled versions, e.g., bigoplus becomes oplus
rm_text	omit text to the left or right of math
long_frac	convert longdiv to frac

Note that rm_fonts and rm_style_syms are implicit in latex_normal, latex_simplified, and latex_list. The long_frac transformation is implicit in latex_simplified and latex_list.

Result object

{
    "detection_list": [],
    "detection_map": {
        "contains_chart": 0,
        "contains_diagram": 0,
        "contains_geometry": 0,
        "contains_graph": 0,
        "contains_table": 0,
        "is_inverted": 0,
        "is_not_math": 0,
        "is_printed": 0
    },
    "latex_normal": "\\lim _ { x \\rightarrow 3 } ( \\frac { x ^ { 2 } + 9 } { x - 3 } )",
    "latex_confidence": 0.86757309488734,
    "latex_confidence_rate": 0.9875550770759583,
    "position": {
        "height": 273,
        "top_left_x": 57,
        "top_left_y": 14,
        "width": 605
    }
}

Field	Type	Description
`text` (optional)	string	Recognized `text` format
`text_display` (optional)	string	Recognized `text_display` format
`latex_normal` (optional)	string	Recognized `latex_normal` format
`latex_simplified` (optional)	string	Recognized `latex_normal` format
`latex_styled` (optional)	string	Recognized `latex_styled` format
`latex_list` (optional)	string[]	Recognized `latex_list` format
`mathml` (optional)	string	Recognized MathML format
`asciimath` (optional)	string	Recognized AsciiMath format
`wolfram` (optional)	string	Recognized Wolfram format
`position` (optional)	object	Position object, pixel coordinates
`detection_list` (optional)	string[]	Detects image properties (see image properties)
`error` (optional)	string	US locale error message
`error_info` (optional)	object	Error info object
`latex_confidence` (optional)	number in [0,1]	Estimated probability 100% correct
`latex_confidence_rate` (optional)	number in [0,1]	Estimated confidence of input quality
`candidates` (optional)	object[]	n_best results
`detected_alphabets` (optional)	[object]	DetectedAlphabet object
`auto_rotate_confidence` (optional)	number in [0,1]	Estimated probability that image needs to be rotated, see Auto rotation
`auto_rotate_degrees` (optional)	number in {0, 90, -90, 180}	Estimated angle of rotation in degrees to put image in correct orientation, see Auto rotation

The detected_alphabets result object contains a field that is true of false for each known alphabet. The field is true if any characters from the alphabet are recognized in the image, regardless of whether any of the result fields contain the characters.

Field	Type	Description
`en`	bool	English
`hi`	bool	Hindi Devenagari
`zh`	bool	Chinese
`ja`	bool	Kana Hiragana or Katakana
`ko`	bool	Hangul Jamo
`ru`	bool	Russian
`th`	bool	Thai

Image properties

The API defines multiple detection types:

Detection	Definition
`contains_diagram`	Contains a diagram.
`is_printed`	The image is taken of printed math, not handwritten math.
`is_not_math`	No valid equation was detected.

Callback request object

Full request body example (includes callback field, shortened src field):

{  
   "src":"data:image/jpeg;base64,...",
   "ocr": ["math", "text"],
   "formats": ["text", "latex_styled"],
   "skip_recrop": true,
   "callback":{  
      "post":"http://RequestBin.com/10z3g561",
      "reply":"integral.jpg"
   }
}

Field	Type	Description
`post`	string	URL to which to make POST callback
`headers`	object	key value pairs of headers to make POST
`reply` (optional)	string	Sets values of `reply` field of callback response object (see callback response object)

Callback response object

{
  "reply": "integral.jpg",
  "result": {
    "detection_list": [
      "is_printed"
    ],
    "detection_map": {
      "contains_chart": 0,
      "contains_diagram": 0,
      "contains_geometry": 0,
      "contains_graph": 0,
      "contains_table": 0,
      "is_inverted": 0,
      "is_not_math": 0,
      "is_printed": 1
    },
    "error": "",
    "latex": "\\int \\frac { 4 x } { \\sqrt { x ^ { 2 } + 1 } } d x",
    "latex_confidence": 0.99817252453161,
    "position": {
      "height": 215,
      "top_left_x": 57,
      "top_left_y": 0,
      "width": 605
    }
  }
}

Field	Type	Description
`reply`	string	Request identifier
`result`	object	Result object

Get OCR results (v3/ocr-results)

Mathpix allows customers to search their results from posts to /v3/text, /v3/strokes, and /v3/latex with a GET request on /v3/ocr-results?search-parameters. The search request must contain a valid app_key header to identify the group owning the results to search. Requests with the metadata improve_mathpix field set to false will not appear in the search results.

Note that this endpoint will only work with API keys created after July 5th, 2020.

Also note that this endpoint will not return OCR if you have the privacy option enabled.

Search parameters

GET https://api.mathpix.com/v3/ocr-results

curl -X GET -H 'app_key: YOUR_APP_KEY' \
    'https://api.mathpix.com/v3/ocr-results?per_page=100&from_date=2020-06-26T03%3A08%3A22.827Z'

Search parameter	Type	Description
`page` (default=1)	number	First page of results to return
`per_page` (default=100)	number	Number of results to return
`from_date` (optional)	string	starting (included) ISO datetime
`to_date` (optional)	string	ending (excluded) ISO datetime
`app_id` (optional)	string	results for the given app_id
`text` (optional)	string	result.text contains the given string
`text_display` (optional)	string	result.text_display contains the given string
`latex_styled` (optional)	string	result.latex_styled contains the given string

Search OCR results

{
    "ocr_results": [
        {
            "timestamp": "2020-06-26T03:08:23.827Z",
            "duration": 0.346,
            "endpoint": "/v3/text",
            "request_args": {
                "formats": [
                    "text"
                ]
            },
            "result": {
                "request_id": "53597c096d7d418e7040072047d7ba25",
                "confidence": 1,
                "confidence_rate": 1,
                "text": "\\( 12+5 x-8=12 x-10 \\)"
            }
        }
    ]
}

Field	Type	Description
`timestamp`	string	ISO timestamp of recorded result information
`endpoint`	object	API endpoint used for upload (eg `/v3/text`, `/v3/strokes`, ...)
`duration`	number	difference between timestamp and when request was received
`request_args`	object	Request body arguments
`result`	object	Result body for request

Error handling

All requests will return error and error_info fields if an argument is missing or incorrect, or if some problem happens while processing the request. The error field contains an en-us string describing the error. The error_info field contains an object providing programmatic information.

Error info object

Field	Type	Description
`id`	string	specifies the error id (see below)
`message`	string	error message
`detail` (optional)	object	Additional error info

Error id strings

Id	Description	Detail fields	HTTP Status
http_unauthorized	Invalid credentials		401
http_max_requests	Too many requests	count	429
json_syntax	JSON syntax error		200
image_missing	Missing URL in request body		200
image_download_error	Error downloading image	url	200
image_decode_error	Cannot decode the image data		200
image_no_content	No content found in image		200
image_not_supported	Image is not math or text		200
image_max_size	Image is too large to process		200
strokes_missing	Missing strokes in request body		200
strokes_syntax_error	Incorrect JSON or strokes format		200
strokes_no_content	No content found in strokes		200
opts_bad_callback	Bad callback field(s)	post?, reply?, batch_id?	200
opts_unknown_ocr	Unknown ocr option(s)	ocr	200
opts_unknown_format	Unknown format option(s)	formats	200
opts_number_required	Option must be a number	name,value	200
opts_value_out_of_range	Value not in accepted range	name,value	200
math_confidence	Low confidence		200
math_syntax	Unrecognized math		200
batch_unknown_id	Unknown batch id	batch_id	200
sys_exception	Server error		200
sys_request_too_large	Max request size is 5mb for images and 512kb for strokes		200

Privacy

Example of a request with the extra privacy setting:

{
   "src":"data:image/jpeg;base64,...",
   "metadata":{
      "improve_mathpix": false
   }
}

By default we make images accessible to our QA team so that we can make improvements.

We also provide an extra privacy option which ensures that no image data or derived information is ever persisted to disk, and no data is available to Mathpix's QA team (we still track the request and how long the request took to complete). Simply add a metadata object to the main request body with the improve_mathpix field set to false (by default it is true). Note that this option means that images and results will not be accessible via our dashboard (accounts.mathpix.com/ocr-api).

Long division

Response for image on the left side:

{
  "detection_map": {
    "contains_chart": 0,
    "contains_diagram": 0,
    "contains_geometry": 0,
    "contains_graph": 0,
    "contains_table": 0,
    "is_inverted": 0,
    "is_not_math": 0,
    "is_printed": 1
  },
  "latex_normal": "8 \\longdiv { 7200 }"
}

We use the special markup \longdiv to represent long division; it is the only nonvalid Latex markup we return. Long division is used much like \sqrt which is visually similar.

Latency considerations

The biggest source of latency is image uploads. The speed of a response from Mathpix API servers is roughly proportional to the size of the image. Try to use images under 100kb for maximum speeds. JPEG compression and image downsizing are recommended to ensure lowest possible latencies.

Vocabulary

Mathpix generates any of the following characters:

!	"	$	&	'
(	)	*	+	,
-	.	/	0	1
2	3	4	5	6
7	8	9	:	;
<	=	>	?	A
B	C	D	E	F
G	H	I	J	K
L	M	N	O	P
Q	R	S	T	U
V	W	X	Y	Z
[	\\	]	^	_
a	b	c	d	e
f	g	h	i	j
k	l	m	n	o
p	q	r	s	t
u	v	w	x	y
z	{	\|	}	~
\\%	\\\\	\\{	\\}
\alpha	\angle	\langle	\rangle	\approx
\because	\begin{array}	\beta	\bot	\cap
\cdot	\cdots	\chi	\circ	\cong
\cup	\dagger	\Delta	\delta	\div
\dot	\dots	\ell	\emptyset	\eta
\end{array}	\epsilon	\equiv	\exists	\kappa
\forall	\frac	\gamma	\Gamma	\geq
\hat	\hbar	\hline	\in	\infty
\int	\Lambda	\lambda	\lceil	\left(
\left.	\left[	\left\\{	\left\|	\Leftrightarrow
\leftrightarrow	\lfloor	\leq	\longdiv	\mathcal
\mp	\mu	\nabla	\neq	\notin
\oint	\Omega	\omega	\operatorna	\oplus
\overline	\otimes	\parallel	\partial	\perp
\Phi	\phi	\pi	\pm	\prime
\prod	\propto	\Psi	\psi	\qquad
\quad	\rceil	\rfloor	\rho	\right)
\right.	\right\}	\right]	\Rightarrow	\rightarrow
\right\|	\sigma	\sim	\simeq	\sqrt
\square	\star	\sum	\subset	\supset
\subseteq	\supseteq	\tau	\text	\therefore
\Theta	\theta	\times	\tilde
\varphi	\vdots	\vec	\wedge	\vee
\Xi	\xi	\zeta