Stay organized with collections Save and categorize content based on your preferences.
With ML Kit's digital ink recognition API, you can recognize handwritten text and classify gestures on a digital surface in hundreds of languages, as well as classify sketches. The digital ink recognition API uses the same technology that powers handwriting recognition in Gboard, Google Translate, and the Quick, Draw! game.
Digital ink recognition allows you to:
Write on the screen instead of typing on a virtual keyboard. This lets users draw characters that are not available on their keyboard, such as ệ, अ or 森 for latin alphabet keyboards.
Perform basic text operations (navigation, editing, selection, and so on) using gestures.
Recognize hand‑drawn shapes and emojis.
Digital ink recognition works with the strokes the user draws on the screen. If you need to read text from images taken with the camera, use the Text Recognition API.
Digital ink recognition works fully offline and is supported on Android and iOS.
Keeps on-device storage low by dynamically downloading language packs as needed
The recognizer takes an Ink object as input. Ink is a vector representation of what the user has written on the screen: a sequence of strokes, each being a list of coordinates with time information called touch points. A stroke starts when the user puts their stylus or finger down and ends when they lift it up. The Ink is passed to a recognizer, which returns one or more possible recognition results, with levels of confidence.
Examples
English handwriting
The image on the left below shows what the user drew on the screen. The image on the right is the corresponding Ink object. It contains the strokes with red dots representing the touch points within each stroke.
There are four strokes. The first two strokes in the Ink object look like this:
Ink
Stroke 1
x
392, 391, 389, 287, ...
y
52, 60, 76, 97, ...
t
0, 37, 56, 75, ...
Stroke 2
x
497, 494, 493, 490, ...
y
167, 165, 165, 165, ...
t
694, 742, 751, 770, ...
...
When you send this Ink to a recognizer for the English language, it returns several possible transcriptions, containing five or six characters. They are ordered by decreasing confidence:
RecognitionResult
RecognitionCandidate #1
handw
RecognitionCandidate #2
handrw
RecognitionCandidate #3
hardw
RecognitionCandidate #4
handu
RecognitionCandidate #5
handwe
Gestures
Gesture classifiers classify an ink stroke into one of nine gesture classes listed below.
Gesture
Example
arch:above arch:below
caret:above caret:below
circle
corner:downleft
scribble
strike
verticalbar
writing
Emoji sketches
The image on the left below shows what the user drew on the screen. The image on the right is the corresponding Ink object. It contains the strokes with red dots representing the touch points within each stroke.
The Ink object contains six strokes.
Ink
Stroke 1
x
269, 266, 262, 255, ...
y
40, 40, 40, 41, ...
t
0, 36, 56, 75, ...
Stroke 2
x
179, 182, 183, 185, ...
y
157, 158, 159, 160, ...
t
2475, 2522, 2531, 2541, ...
...
When you send this Ink to the emoji recognizer, you get several possible transcriptions, ordered by decreasing confidence:
[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Missing the information I need","missingTheInformationINeed","thumb-down"],["Too complicated / too many steps","tooComplicatedTooManySteps","thumb-down"],["Out of date","outOfDate","thumb-down"],["Samples / code issue","samplesCodeIssue","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2024-07-10 UTC."],[[["\u003cp\u003eML Kit's Digital Ink Recognition API recognizes handwritten text and gestures, converting them into digital format, comparable to the technology used in Gboard and Google Translate.\u003c/p\u003e\n"],["\u003cp\u003eThis API enables on-screen writing in various languages, using gestures for text editing, and recognizing hand-drawn shapes and emojis, all without an internet connection.\u003c/p\u003e\n"],["\u003cp\u003eIt supports over 300 languages and 25+ writing systems, along with gesture classification and emoji recognition, functioning by processing stroke data of user input.\u003c/p\u003e\n"],["\u003cp\u003eDevelopers can integrate this feature to allow users to write with styluses or fingers, replacing or supplementing traditional keyboard input for a more natural and versatile user experience.\u003c/p\u003e\n"]]],["ML Kit's digital ink recognition API converts handwritten text, gestures, and sketches into digital formats. It operates offline on Android and iOS, supporting 300+ languages and 25+ writing systems. The API processes user-drawn strokes (Ink objects) to recognize text, emojis, and basic shapes, returning ranked recognition results. Gestures are classified into nine categories, aiding in text operations and user interface actions. Language packs are dynamically downloaded for space efficiency.\n"],null,["With ML Kit's digital ink recognition API, you can recognize handwritten text\nand classify gestures on a digital surface in hundreds of languages, as well as\nclassify sketches. The digital ink recognition API uses the same technology that\npowers handwriting recognition in Gboard, Google Translate, and the\n[Quick, Draw!](https://quickdraw.withgoogle.com/) game.\n\nDigital ink recognition allows you to:\n\n- Write on the screen instead of typing on a virtual keyboard. This lets users draw characters that are not available on their keyboard, such as ệ, अ or 森 for latin alphabet keyboards.\n- Perform basic text operations (navigation, editing, selection, and so on) using gestures.\n- Recognize hand‑drawn shapes and emojis.\n\nDigital ink recognition works with the strokes the user draws on the screen. If\nyou need to read text from images taken with the camera, use the\n[Text Recognition API](/ml-kit/vision/text-recognition).\n\nDigital ink recognition works fully offline and is supported on Android and iOS.\n\n[iOS](/ml-kit/vision/digital-ink-recognition/ios)\n[Android](/ml-kit/vision/digital-ink-recognition/android)\n\nKey Capabilities\n\n- Converts handwritten text to sequences of unicode characters\n- Runs on the device in near real time\n- The user's handwriting stays on the device, recognition is performed without any network connection\n- Supports 300+ languages and 25+ writing systems, see the [complete list of supported languages](/ml-kit/vision/digital-ink-recognition/base-models#text)\n - Supports gesture classification for these languages via [`-x-gesture` extensions](/ml-kit/vision/digital-ink-recognition/base-models#text)\n- Recognizes emojis and basic shapes\n- Keeps on-device storage low by dynamically downloading language packs as needed\n\nThe recognizer takes an `Ink` object as input. `Ink` is a vector representation\nof what the user has written on the screen: a sequence of *strokes* , each being\na list of coordinates with time information called *touch points* . A stroke\nstarts when the user puts their stylus or finger down and ends when they lift it\nup. The `Ink` is passed to a recognizer, which returns one or more possible\nrecognition results, with levels of confidence.\n\nExamples\n\nEnglish handwriting\n\nThe image on the left below shows what the user drew on the screen. The image on\nthe right is the corresponding `Ink` object. It contains the strokes with red\ndots representing the touch points within each stroke.\n\n\nThere are four strokes. The first two strokes in the `Ink` object look like\nthis:\n\n| **Ink** |||\n|----------|-----|-------------------------|\n| Stroke 1 | `x` | 392, 391, 389, 287, ... |\n| Stroke 1 | `y` | 52, 60, 76, 97, ... |\n| Stroke 1 | `t` | 0, 37, 56, 75, ... |\n| Stroke 2 | `x` | 497, 494, 493, 490, ... |\n| Stroke 2 | `y` | 167, 165, 165, 165, ... |\n| Stroke 2 | `t` | 694, 742, 751, 770, ... |\n| ... | | |\n\nWhen you send this `Ink` to a recognizer for the English language, it returns\nseveral possible transcriptions, containing five or six characters. They are\nordered by decreasing confidence:\n\n| **RecognitionResult** ||\n|-------------------------|--------|\n| RecognitionCandidate #1 | handw |\n| RecognitionCandidate #2 | handrw |\n| RecognitionCandidate #3 | hardw |\n| RecognitionCandidate #4 | handu |\n| RecognitionCandidate #5 | handwe |\n\nGestures\n\nGesture classifiers classify an ink stroke into one of nine gesture classes\nlisted below.\n\n| Gesture | Example |\n|-----------------------------|---------|\n| `arch:above` `arch:below` | |\n| `caret:above` `caret:below` | |\n| `circle` | |\n| corner:downleft | |\n| `scribble` | |\n| `strike` | |\n| `verticalbar` | |\n| `writing` | |\n\n| **Note:** It is not always possible to reliably distinguish some gestures from writing. For example, the `verticalbar` gesture may look exactly like the digit `1` or letter `l` when they are written as a vertical lines. To allow the user to use both gestures and writing, your application may need to consider the position of the writing or gesture: for the writing over existing text, prefer the gesture interpretation; for the writing over empty space, prefer the text interpretation.\n\nEmoji sketches\n\nThe image on the left below shows what the user drew on the screen. The image on\nthe right is the corresponding `Ink` object. It contains the strokes with red\ndots representing the touch points within each stroke.\n\n\nThe `Ink` object contains six strokes.\n\n\n| **Ink** |||\n|----------|-----|-----------------------------|\n| Stroke 1 | `x` | 269, 266, 262, 255, ... |\n| Stroke 1 | `y` | 40, 40, 40, 41, ... |\n| Stroke 1 | `t` | 0, 36, 56, 75, ... |\n| Stroke 2 | `x` | 179, 182, 183, 185, ... |\n| Stroke 2 | `y` | 157, 158, 159, 160, ... |\n| Stroke 2 | `t` | 2475, 2522, 2531, 2541, ... |\n| ... | | |\n\nWhen you send this `Ink` to the emoji recognizer, you get several possible\ntranscriptions, ordered by decreasing confidence:\n\n| **RecognitionResult** ||\n|-------------------------|--------------|\n| RecognitionCandidate #1 | 😂 (U+1f62d) |\n| RecognitionCandidate #2 | 😅 (U+1f605) |\n| RecognitionCandidate #3 | 😹 (U+1f639) |\n| RecognitionCandidate #4 | 😄 (U+1f604) |\n| RecognitionCandidate #5 | 😆 (U+1f606) |"]]
