Android项目集成Google ML OCR Tesseract出现识别乱码问题求助

我在Android项目中集成Google ML OCR Tesseract时，拍摄图片后识别出乱码文本（如m”;-.v__m-~_;;;;;;§'F"WWW If?），以下是实现代码，请求排查问题所在：

class MainActivity : ComponentActivity() {
private lateinit var tessBaseAPI: TessBaseAPI
private lateinit var imageView: ImageView
private lateinit var textView: TextView
private val requestImageCapture = 1
private val takePicture =
registerForActivityResult(ActivityResultContracts.StartActivityForResult()) { result ->

        if (result.resultCode == RESULT_OK) {
            val resultInfo = result.data?.extras?.get("data")
            val imageBitmap = result.data?.extras?.get("data") as Bitmap?
            imageBitmap?.let {
                imageView.setImageBitmap(it)
                recognizeText(it)
            }
        }
    }

override fun onCreate(savedInstanceState: Bundle?) {
    super.onCreate(savedInstanceState)
    setContentView(R.layout.activity_main)

    imageView = findViewById(R.id.imageView)
    textView = findViewById(R.id.textView)
    val captureButton: Button = findViewById(R.id.captureButton)

    copyLanguageDataToInternalStorage()
    initTesseract()

    captureButton.setOnClickListener {
        dispatchTakePictureIntent()
    }
}

private fun initTesseract() {
    val datapath = "${filesDir}/tesseract/"
    val dir = File(datapath)
    if (!dir.exists()) {
        dir.mkdirs()
    }
    val datafilepath = "${datapath}/tessdata/"
    val filesDir = File(datafilepath)
    if (!filesDir.exists()) {
        filesDir.mkdirs()
    }
    Log.d("dataPath", "data: $datapath")
    tessBaseAPI = TessBaseAPI()
    tessBaseAPI.init(datapath, "eng")
}

private fun copyLanguageDataToInternalStorage() {
    Log.d("test", "testing")
    try {
        val assetManager = assets
        val inputStream = assetManager.open("tesseract/tessdata/eng.traineddata")
        val outputStream =
            FileOutputStream(File(filesDir, "tesseract/tessdata/eng.traineddata"))

        inputStream.use { input ->
            outputStream.use { output ->
                input.copyTo(output)
            }
        }

    } catch (e: IOException) {

        e.printStackTrace()
    }
}

private fun dispatchTakePictureIntent() {
    val takePictureIntent = Intent(MediaStore.ACTION_IMAGE_CAPTURE)
    if (takePictureIntent.resolveActivity(packageManager) != null) {
        takePicture.launch(takePictureIntent)
    }
}

private fun recognizeText(bitmap: Bitmap) {
    tessBaseAPI.setImage(bitmap)
    val recognizedText = tessBaseAPI.getUTF8Text()
    textView.text = recognizedText
}

override fun onDestroy() {
    super.onDestroy()
    tessBaseAPI.end()
}

核心问题排查与修复方案 #

• 拍摄图片分辨率过低
  MediaStore.ACTION_IMAGE_CAPTURE返回的"data"字段只是缩略图，分辨率极低，Tesseract无法从中提取有效文字。必须获取完整尺寸的图片：

  1. 先创建临时文件保存拍摄的图片
  2. 通过MediaStore.EXTRA_OUTPUT指定图片保存路径
  3. 拍摄完成后从该路径加载完整Bitmap

  修改后的dispatchTakePictureIntent及相关代码示例：

  private lateinit var photoFile: File
  
  private fun dispatchTakePictureIntent() {
      val takePictureIntent = Intent(MediaStore.ACTION_IMAGE_CAPTURE)
      photoFile = createImageFile()
      photoFile?.also {
          val photoURI = FileProvider.getUriForFile(
              this,
              "${packageName}.fileprovider",
              it
          )
          takePictureIntent.putExtra(MediaStore.EXTRA_OUTPUT, photoURI)
          takePictureIntent.resolveActivity(packageManager)?.let {
              takePicture.launch(takePictureIntent)
          }
      }
  }
  
  private fun createImageFile(): File {
      val timeStamp: String = SimpleDateFormat("yyyyMMdd_HHmmss", Locale.getDefault()).format(Date())
      val storageDir: File? = getExternalFilesDir(Environment.DIRECTORY_PICTURES)
      return File.createTempFile(
          "JPEG_${timeStamp}_",
          ".jpg",
          storageDir
      )
  }
  
  // 修改回调逻辑加载完整图片
  private val takePicture = registerForActivityResult(ActivityResultContracts.StartActivityForResult()) { result ->
      if (result.resultCode == RESULT_OK) {
          val fullBitmap = BitmapFactory.decodeFile(photoFile.absolutePath)
          imageView.setImageBitmap(fullBitmap)
          recognizeText(fullBitmap)
      }
  }
  

• 图片未做预处理
  Tesseract对输入图片要求较高，彩色图、低对比度图片都会导致识别失败。需要将Bitmap转为灰度图并做二值化处理：

  private fun preprocessBitmap(bitmap: Bitmap): Bitmap {
      // 转为灰度图
      val grayBitmap = Bitmap.createBitmap(bitmap.width, bitmap.height, Bitmap.Config.ARGB_8888)
      val canvas = Canvas(grayBitmap)
      val paint = Paint()
      val colorMatrix = ColorMatrix()
      colorMatrix.setSaturation(0f)
      val filter = ColorMatrixColorFilter(colorMatrix)
      paint.colorFilter = filter
      canvas.drawBitmap(bitmap, 0f, 0f, paint)
  
      // 二值化处理
      val threshold = 127
      val pixels = IntArray(grayBitmap.width * grayBitmap.height)
      grayBitmap.getPixels(pixels, 0, grayBitmap.width, 0, 0, grayBitmap.width, grayBitmap.height)
      for (i in pixels.indices) {
          val gray = (pixels[i] and 0xff)
          pixels[i] = if (gray > threshold) 0xffffffff.toInt() else 0xff000000.toInt()
      }
      grayBitmap.setPixels(pixels, 0, grayBitmap.width, 0, 0, grayBitmap.width, grayBitmap.height)
      return grayBitmap
  }
  

  在recognizeText中调用预处理：

  private fun recognizeText(bitmap: Bitmap) {
      val processedBitmap = preprocessBitmap(bitmap)
      tessBaseAPI.setImage(processedBitmap)
      val recognizedText = tessBaseAPI.getUTF8Text()
      textView.text = recognizedText
      processedBitmap.recycle() // 释放资源
  }
  

• 语言包复制与初始化问题

  1. 检查eng.traineddata是否正确放置在assets/tesseract/tessdata/目录下
  2. 复制完成后校验文件大小，避免重复复制或复制不完整：

     private fun copyLanguageDataToInternalStorage() {
         try {
             val assetManager = assets
             val destFile = File(filesDir, "tesseract/tessdata/eng.traineddata")
             // 如果文件已存在且大小一致，跳过复制
             if (destFile.exists()) {
                 val assetFd = assetManager.openFd("tesseract/tessdata/eng.traineddata")
                 if (destFile.length() == assetFd.length) {
                     assetFd.close()
                     return
                 }
                 assetFd.close()
             }
             val inputStream = assetManager.open("tesseract/tessdata/eng.traineddata")
             val outputStream = FileOutputStream(destFile)
             inputStream.use { input ->
                 outputStream.use { output ->
                     input.copyTo(output)
                 }
             }
         } catch (e: IOException) {
             e.printStackTrace()
         }
     }
     

  3. 确认Tesseract初始化成功，添加日志校验：

     private fun initTesseract() {
         val datapath = "${filesDir}/tesseract/"
         val tessDataFile = File("${datapath}/tessdata/eng.traineddata")
         if (!tessDataFile.exists()) {
             Log.e("Tesseract", "语言包文件不存在")
             return
         }
         val dir = File(datapath)
         if (!dir.exists()) dir.mkdirs()
         val filesDir = File("${datapath}/tessdata/")
         if (!filesDir.exists()) filesDir.mkdirs()
         tessBaseAPI = TessBaseAPI()
         val initSuccess = tessBaseAPI.init(datapath, "eng")
         Log.d("Tesseract", "初始化结果: $initSuccess")
     }
     

内容的提问来源于stack exchange，提问作者Vijay Chandan Kandadai