android - 如何处理firebase ML文本识别返回的文本

Question

我正在使用 firebase ML vision 从图像中读取 RC 细节。

implementation "com.google.android.gms:play-services-mlkit-text-recognition:17.0.1"


private fun Activity.recognizeText(imageSting: String) {
    var image: InputImage? = null
    try {
        image = InputImage.fromFilePath(getContext(), Uri.fromFile(File(imageSting)))
    } catch (e: IOException) {
        e.printStackTrace()
    }
    // [START get_detector_default]
    val recognizer = TextRecognition.getClient(TextRecognizerOptions.DEFAULT_OPTIONS)
    // [END get_detector_default]

    // [START run_detector]
    val result = recognizer.process(image)
        .addOnSuccessListener { visionText ->
           processTextBlock(visionText)
       
        }
        .addOnFailureListener { e ->
          
            Log.d("test", e.localizedMessage);
        }
}

它工作正常，文本返回成功。

数据列表通过 cornerPoints 和 boundingBox 返回， map 数据如何正确..前注册。有实际值(value)的编号

Government of 1Tamil Nadu
Certificate of Registration
O
Reg. No.
Date of Reg.
TNO6P7094
30-06-2015
Reg. Valid Till
29-06-2030
Chassis No.
ME4JF504FFT459059
Engine No.
JF50ET 2460128
Owner
02
Sr. No.
Owner Name
NIKIL KUMAR ROY
Fuel Used
PETROL
Son/Daughter/Wife of
MANIK ROY
Address
NO 456/HPTRAILWAY
COLONY 4TH STREET
AYANAVARAM Chennai TN 600023

Answer 1

好吧，如果你想要 reg.no。你可以采用多种方式:

1. 通过正则表达式从所有文本中过滤掉

val regex = Regex("[A-Z]{3}[0-9]{1}[A-Z]{1}[0-9]{4}")
text.textBlocks.map {
   it.lines.filter { 
     regex.matches(it.text)
   }
}

1. 从图片区域过滤掉(可能然后使用正则表达式)

// this is camera preview
previewCamera.getGlobalVisibleRect(previewCameraRect)
// here you would calculate the position on the picture, where you would expect needed thing. Do not forget to calculate on change of rotation
Rect(left.roundToInt(), top.roundToInt(), right.roundToInt(), bottom.roundToInt())

text.textBlocks.filter { textBlock ->
   textBlock.boundingBox?.let { rect.contains(it) } ?: false 
}

1. 这两者的结合。您可能可以制作更精细的算法，将位置和文本匹配结合起来。所以像

// Pseudocode
foreach (textBlocks as textBlock) {
   if (textBlock.text.equals("Reg.No")) {
      position = textBlock.boundingBox
      continue
   }

   if (isNear(textBlock.boundingBox, position) && regex.matches(textBlock.text)) {
     // this is part of text you are looking for
   }
}