Join us for Firebase Summit on November 10, 2021. Tune in to learn how Firebase can help you accelerate app development, release with confidence, and scale with ease. Register

Recognize Text in Images Securely with Cloud Vision using Firebase Auth and Functions on Android

In order to call a Google Cloud API from your app, you need to create an intermediate REST API that handles authorization and protects secret values such as API keys. You then need to write code in your mobile app to authenticate to and communicate with this intermediate service.

One way to create this REST API is by using Firebase Authentication and Functions, which gives you a managed, serverless gateway to Google Cloud APIs that handles authentication and can be called from your mobile app with pre-built SDKs.

This guide demonstrates how to use this technique to call the Cloud Vision API from your app. This method will allow all authenticated users to access Cloud Vision billed services through your Cloud project, so consider whether this auth mechanism is sufficient for your use case before proceeding.

Before you begin

Configure your project

  1. If you haven't already, add Firebase to your Android project.
  2. If you have not already enabled Cloud-based APIs for your project, do so now:

    1. Open the Firebase ML APIs page of the Firebase console.
    2. If you have not already upgraded your project to the Blaze pricing plan, click Upgrade to do so. (You will be prompted to upgrade only if your project isn't on the Blaze plan.)

      Only Blaze-level projects can use Cloud-based APIs.

    3. If Cloud-based APIs aren't already enabled, click Enable Cloud-based APIs.
  3. Configure your existing Firebase API keys to disallow access to the Cloud Vision API:
    1. Open the Credentials page of the Cloud console.
    2. For each API key in the list, open the editing view, and in the Key Restrictions section, add all of the available APIs except the Cloud Vision API to the list.

Deploy the callable function

Next, deploy the Cloud Function you will use to bridge your app and the Cloud Vision API. The functions-samples repository contains an example you can use.

By default, accessing the Cloud Vision API through this function will allow only authenticated users of your app access to the Cloud Vision API. You can modify the function for different requirements.

To deploy the function:

  1. Clone or download the functions-samples repo and change to the vision-annotate-image directory:
    git clone https://github.com/firebase/functions-samples
    cd vision-annotate-image
    
  2. Install dependencies:
    cd functions
    npm install
    cd ..
    
  3. If you don't have the Firebase CLI, install it.
  4. Initialize a Firebase project in the vision-annotate-image directory. When prompted, select your project in the list.
    firebase init
  5. Deploy the function:
    firebase deploy --only functions:annotateImage

Add Firebase Auth to your app

The callable function deployed above will reject any request from non-authenticated users of your app. If you have not already done so, you will need to add Firebase Auth to your app.

Add necessary dependencies to your app

  • Add the dependencies for the Firebase Functions and gson Android libraries to your module (app-level) Gradle file (usually app/build.gradle):
    implementation 'com.google.firebase:firebase-functions:20.0.1'
    implementation 'com.google.code.gson:gson:2.8.6'
    
  • Now you are ready to start recognizing text in images.

    1. Prepare the input image

    In order to call Cloud Vision, the image must be formatted as a base64-encoded string. To process an image from a saved file URI:
    1. Get the image as a Bitmap object:

      Java

      Bitmap bitmap = MediaStore.Images.Media.getBitmap(getContentResolver(), uri);

      Kotlin+KTX

      var bitmap: Bitmap = MediaStore.Images.Media.getBitmap(contentResolver, uri)
    2. Optionally, scale down the image to save on bandwidth. See the Cloud Vision recommended image sizes.

      Java

      private Bitmap scaleBitmapDown(Bitmap bitmap, int maxDimension) {
          int originalWidth = bitmap.getWidth();
          int originalHeight = bitmap.getHeight();
          int resizedWidth = maxDimension;
          int resizedHeight = maxDimension;
      
          if (originalHeight > originalWidth) {
              resizedHeight = maxDimension;
              resizedWidth = (int) (resizedHeight * (float) originalWidth / (float) originalHeight);
          } else if (originalWidth > originalHeight) {
              resizedWidth = maxDimension;
              resizedHeight = (int) (resizedWidth * (float) originalHeight / (float) originalWidth);
          } else if (originalHeight == originalWidth) {
              resizedHeight = maxDimension;
              resizedWidth = maxDimension;
          }
          return Bitmap.createScaledBitmap(bitmap, resizedWidth, resizedHeight, false);
      }

      Kotlin+KTX

      private fun scaleBitmapDown(bitmap: Bitmap, maxDimension: Int): Bitmap {
          val originalWidth = bitmap.width
          val originalHeight = bitmap.height
          var resizedWidth = maxDimension
          var resizedHeight = maxDimension
          if (originalHeight > originalWidth) {
              resizedHeight = maxDimension
              resizedWidth =
                      (resizedHeight * originalWidth.toFloat() / originalHeight.toFloat()).toInt()
          } else if (originalWidth > originalHeight) {
              resizedWidth = maxDimension
              resizedHeight =
                      (resizedWidth * originalHeight.toFloat() / originalWidth.toFloat()).toInt()
          } else if (originalHeight == originalWidth) {
              resizedHeight = maxDimension
              resizedWidth = maxDimension
          }
          return Bitmap.createScaledBitmap(bitmap, resizedWidth, resizedHeight, false)
      }

      Java

      // Scale down bitmap size
      bitmap = scaleBitmapDown(bitmap, 640);

      Kotlin+KTX

      // Scale down bitmap size
      bitmap = scaleBitmapDown(bitmap, 640)
    3. Convert the bitmap object to a base64 encoded string:

      Java

      // Convert bitmap to base64 encoded string
      ByteArrayOutputStream byteArrayOutputStream = new ByteArrayOutputStream();
      bitmap.compress(Bitmap.CompressFormat.JPEG, 100, byteArrayOutputStream);
      byte[] imageBytes = byteArrayOutputStream.toByteArray();
      String base64encoded = Base64.encodeToString(imageBytes, Base64.NO_WRAP);

      Kotlin+KTX

      // Convert bitmap to base64 encoded string
      val byteArrayOutputStream = ByteArrayOutputStream()
      bitmap.compress(Bitmap.CompressFormat.JPEG, 100, byteArrayOutputStream)
      val imageBytes: ByteArray = byteArrayOutputStream.toByteArray()
      val base64encoded = Base64.encodeToString(imageBytes, Base64.NO_WRAP)
    4. The image represented by the Bitmap object must be upright, with no additional rotation required.

    2. Invoke the callable function to recognize text

    To recognize text in an image, invoke the callable function, passing a JSON Cloud Vision request.

    1. First, initialize an instance of Cloud Functions:

      Java

      private FirebaseFunctions mFunctions;
      // ...
      mFunctions = FirebaseFunctions.getInstance();
      

      Kotlin+KTX

      private lateinit var functions: FirebaseFunctions
      // ...
      functions = Firebase.functions
      
    2. Define a method for invoking the function:

      Java

      private Task<JsonElement> annotateImage(String requestJson) {
          return mFunctions
                  .getHttpsCallable("annotateImage")
                  .call(requestJson)
                  .continueWith(new Continuation<HttpsCallableResult, JsonElement>() {
                      @Override
                      public JsonElement then(@NonNull Task<HttpsCallableResult> task) {
                          // This continuation runs on either success or failure, but if the task
                          // has failed then getResult() will throw an Exception which will be
                          // propagated down.
                          return JsonParser.parseString(new Gson().toJson(task.getResult().getData()));
                      }
                  });
      }
      

      Kotlin+KTX

      private fun annotateImage(requestJson: String): Task<JsonElement> {
          return functions
                  .getHttpsCallable("annotateImage")
                  .call(requestJson)
                  .continueWith { task ->
                      // This continuation runs on either success or failure, but if the task
                      // has failed then result will throw an Exception which will be
                      // propagated down.
                      val result = task.result?.data
                      JsonParser.parseString(Gson().toJson(result))
                  }
      }
      
    3. Create the JSON request. The Cloud Vision API supports two Types of text detection: TEXT_DETECTION and DOCUMENT_TEXT_DETECTION. See the Cloud Vision OCR Docs for the difference between the two use cases.

      Java

      // Create json request to cloud vision
      JsonObject request = new JsonObject();
      // Add image to request
      JsonObject image = new JsonObject();
      image.add("content", new JsonPrimitive(base64encoded));
      request.add("image", image);
      //Add features to the request
      JsonObject feature = new JsonObject();
      feature.add("type", new JsonPrimitive("TEXT_DETECTION"));
      // Alternatively, for DOCUMENT_TEXT_DETECTION:
      //feature.add("type", new JsonPrimitive("DOCUMENT_TEXT_DETECTION"));
      JsonArray features = new JsonArray();
      features.add(feature);
      request.add("features", features);
      

      Kotlin+KTX

      // Create json request to cloud vision
      val request = JsonObject()
      // Add image to request
      val image = JsonObject()
      image.add("content", JsonPrimitive(base64encoded))
      request.add("image", image)
      //Add features to the request
      val feature = JsonObject()
      feature.add("type", JsonPrimitive("TEXT_DETECTION"))
      // Alternatively, for DOCUMENT_TEXT_DETECTION:
      // feature.add("type", JsonPrimitive("DOCUMENT_TEXT_DETECTION"))
      val features = JsonArray()
      features.add(feature)
      request.add("features", features)
      

      Optionally, provide language hints to assist with language detection (see supported languages):

      Java

      JsonObject imageContext = new JsonObject();
      JsonArray languageHints = new JsonArray();
      languageHints.add("en");
      imageContext.add("languageHints", languageHints);
      request.add("imageContext", imageContext);
      

      Kotlin+KTX

      val imageContext = JsonObject()
      val languageHints = JsonArray()
      languageHints.add("en")
      imageContext.add("languageHints", languageHints)
      request.add("imageContext", imageContext)
      
    4. Finally, invoke the function:

      Java

      annotateImage(request.toString())
              .addOnCompleteListener(new OnCompleteListener<JsonElement>() {
                  @Override
                  public void onComplete(@NonNull Task<JsonElement> task) {
                      if (!task.isSuccessful()) {
                          // Task failed with an exception
                          // ...
                      } else {
                          // Task completed successfully
                          // ...
                      }
                  }
              });
      

      Kotlin+KTX

      annotateImage(request.toString())
              .addOnCompleteListener { task ->
                  if (!task.isSuccessful) {
                      // Task failed with an exception
                      // ...
                  } else {
                      // Task completed successfully
                      // ...
                  }
              }
      

    3. Extract text from blocks of recognized text

    If the text recognition operation succeeds, a JSON response of BatchAnnotateImagesResponse will be returned in the task's result. The text annotations can be found in the fullTextAnnotation object.

    You can get the recognized text as a string in the text field. For example:

    Java

    JsonObject annotation = task.getResult().getAsJsonArray().get(0).getAsJsonObject().get("fullTextAnnotation").getAsJsonObject();
    System.out.format("%nComplete annotation:%n");
    System.out.format("%s%n", annotation.get("text").getAsString());
    

    Kotlin+KTX

    val annotation = task.result!!.asJsonArray[0].asJsonObject["fullTextAnnotation"].asJsonObject
    System.out.format("%nComplete annotation:")
    System.out.format("%n%s", annotation["text"].asString)
    

    You can also get information specific to regions of the image. For each block, paragraph, word, and symbol, you can get the text recognized in the region and the bounding coordinates of the region. For example:

    Java

    for (JsonElement page : annotation.get("pages").getAsJsonArray()) {
        StringBuilder pageText = new StringBuilder();
        for (JsonElement block : page.getAsJsonObject().get("blocks").getAsJsonArray()) {
            StringBuilder blockText = new StringBuilder();
            for (JsonElement para : block.getAsJsonObject().get("paragraphs").getAsJsonArray()) {
                StringBuilder paraText = new StringBuilder();
                for (JsonElement word : para.getAsJsonObject().get("words").getAsJsonArray()) {
                    StringBuilder wordText = new StringBuilder();
                    for (JsonElement symbol : word.getAsJsonObject().get("symbols").getAsJsonArray()) {
                        wordText.append(symbol.getAsJsonObject().get("text").getAsString());
                        System.out.format("Symbol text: %s (confidence: %f)%n", symbol.getAsJsonObject().get("text").getAsString(), symbol.getAsJsonObject().get("confidence").getAsFloat());
                    }
                    System.out.format("Word text: %s (confidence: %f)%n%n", wordText.toString(), word.getAsJsonObject().get("confidence").getAsFloat());
                    System.out.format("Word bounding box: %s%n", word.getAsJsonObject().get("boundingBox"));
                    paraText.append(wordText.toString()).append(" ");
                }
                System.out.format("%nParagraph:%n%s%n", paraText);
                System.out.format("Paragraph bounding box: %s%n", para.getAsJsonObject().get("boundingBox"));
                System.out.format("Paragraph Confidence: %f%n", para.getAsJsonObject().get("confidence").getAsFloat());
                blockText.append(paraText);
            }
            pageText.append(blockText);
        }
    }
    

    Kotlin+KTX

    for (page in annotation["pages"].asJsonArray) {
        var pageText = ""
        for (block in page.asJsonObject["blocks"].asJsonArray) {
            var blockText = ""
            for (para in block.asJsonObject["paragraphs"].asJsonArray) {
                var paraText = ""
                for (word in para.asJsonObject["words"].asJsonArray) {
                    var wordText = ""
                    for (symbol in word.asJsonObject["symbols"].asJsonArray) {
                        wordText += symbol.asJsonObject["text"].asString
                        System.out.format("Symbol text: %s (confidence: %f)%n",
                            symbol.asJsonObject["text"].asString, symbol.asJsonObject["confidence"].asFloat)
                    }
                    System.out.format("Word text: %s (confidence: %f)%n%n", wordText,
                        word.asJsonObject["confidence"].asFloat)
                    System.out.format("Word bounding box: %s%n", word.asJsonObject["boundingBox"])
                    paraText = String.format("%s%s ", paraText, wordText)
                }
                System.out.format("%nParagraph: %n%s%n", paraText)
                System.out.format("Paragraph bounding box: %s%n", para.asJsonObject["boundingBox"])
                System.out.format("Paragraph Confidence: %f%n", para.asJsonObject["confidence"].asFloat)
                blockText += paraText
            }
            pageText += blockText
        }
    }