Reference: https://www.raywenderlich.com/7479-text-recognition-with-ml-kit

MLKitTextDetector.kt

fun runTextRecognition(bitmap: Bitmap) {
  val image = FirebaseVisionImage.fromBitmap(bitmap)
  // FirebaseVisionTextRecognizer will be used
  val detector = FirebaseVision.getInstance().visionTextDetector
  detector.detectInImage(image)
    .addOnSuccessListener {
      texts ->
      val blocks = texts.blocks
      if (blocks.size == 0) { /*nothing found*/ return }
      blocks.forEach {
        block -> // a bunch of text
        block.lines.forEach {
          line -> // a line of text
          line.elements.forEach {
            element -> // each token
            element.text // OCR result
            element.boundingBox // where it is in image
          }
        }
      }
    }
    .addOnFailureListener {
      e -> e.printStackTrace()
    } 
}
/*
If using cloud detector instead, processing would be similar, 
because we are using FirebaseVisionDocumentTextRecognizer,
structure of the returned result would be:

pages -> blocks -> words -> symbols

So we'll need to join symbols together back to a word:
*/
fun toString(word: FirebaseVisionCloudText.Word): String
  = word.symbols.joinToString("") { it.text }
/*
Notice: @ symbol will be a separate word, that's why the author joined
2 words together to be a single token for the regex matching function.
That's unnecessary for other use cases, so only zipWithNext if needed.

Why cloud? It should have higher accuracy, but cost you money, so nah.
*/

最终效果：

MLKit 的组成：