Android CameraX 图像旋转

Posted 2023-05-07

【中文标题】Android CameraX 图像旋转

【英文标题】：Android CameraX image rotated

【发布时间】：2021-01-03 08:33:54

【问题描述】：

我已关注 Google CameraX code lab 实现自定义摄像头。相机预览很好，但是当我在图像捕获图像旋转后拍摄图像时。我正在以纵向模式拍摄图像，但保存的图像是横向的。这是配置相机的方法

private fun startCamera() 

    val cameraProviderFuture = ProcessCameraProvider.getInstance(this)

    cameraProviderFuture.addListener(Runnable 
        // Used to bind the lifecycle of cameras to the lifecycle owner
        val cameraProvider: ProcessCameraProvider = cameraProviderFuture.get()

        // Preview
        val preview = Preview.Builder()
            .setTargetRotation(this.windowManager.defaultDisplay.rotation)
            .build()
            .also 
                it.setSurfaceProvider(viewFinder.createSurfaceProvider())
            

        imageCapture = ImageCapture.Builder()
            .setTargetRotation(this.windowManager.defaultDisplay.rotation)
            .build()

        val imageAnalyzer = ImageAnalysis.Builder()
            .build()
            .also 
                it.setAnalyzer(cameraExecutor, LuminosityAnalyzer  luma ->
                    Log.d(TAG, "Average luminosity: $luma")
                )
            

        // Select back camera as a default
        val cameraSelector = CameraSelector.DEFAULT_BACK_CAMERA

        try 
            // Unbind use cases before rebinding
            cameraProvider.unbindAll()

            // Bind use cases to camera
            cameraProvider.bindToLifecycle(
                this, cameraSelector, preview, imageCapture, imageAnalyzer)

         catch(exc: Exception) 
            Log.e(TAG, "Use case binding failed", exc)
        

    , ContextCompat.getMainExecutor(this))

以下是获取图像的方法：

private fun takePhoto() 
    val imageCapture = imageCapture ?: return

    // Create time-stamped output file to hold the image
    val photoFile = File(
        outputDirectory,
        SimpleDateFormat(FILENAME_FORMAT, Locale.US
        ).format(System.currentTimeMillis()) + ".jpg")

    // Create output options object which contains file + metadata
    val outputOptions = ImageCapture.OutputFileOptions.Builder(photoFile).build()

    // Set up image capture listener, which is triggered after photo has
    // been taken
    imageCapture.takePicture(
        outputOptions, ContextCompat.getMainExecutor(this), object : ImageCapture.OnImageSavedCallback 
            override fun onError(exc: ImageCaptureException) 
                Log.e(TAG, "Photo capture failed: $exc.message", exc)
            

            override fun onImageSaved(output: ImageCapture.OutputFileResults) 
                val savedUri = Uri.fromFile(photoFile)
                val msg = "Photo capture succeeded: $savedUri"
                val bitmap = MediaStore.Images.Media.getBitmap(contentResolver, savedUri)
                ivCapturedImage.setImageBitmap(bitmap)
                setCaptureUI(false)
                Log.d(TAG, msg)
            
        )

使用EXIF拍摄后是否需要自己旋转图像，或者我可以在配置相机时修复它？

【问题讨论】：

澄清问题：您使用的是三星设备吗？他们中的许多人都有一个已知的错误，即在保存时以错误的方向记录图像。

@JohnLord 你有这个声明的来源吗？是否有任何已知的解决方法？

唯一已知的解决方法是保存图像，然后读取 exif 数据。这是一个众所周知的问题，并且在 *** 上有各种关于它的帖子，例如这个。 ***.com/questions/47261434/…我们公司有数百个三星平板电脑，我们必须在上面的链接中包含一个类似的修复程序，尽管我们的更简单，因为我们的平板电脑被锁定为纵向。在上面的链接中，他们将读取的 exif 数据与当前设备方向进行比较。

【参考方案1】：

默认情况下，ImageCapture 将捕获的方向设置为显示旋转。如果图像保存到磁盘，旋转将在 EXIF 中。

您的设备是否处于锁定纵向模式？在这种情况下，显示旋转与设备的方向不匹配，您需要自己设置目标旋转。示例。

// The value is whatever the display rotation should be, if the device orientation is not locked.
imageCapture.setTargetRotation(...) 

或者，您可以简单地使用LifecycleCameraController API。它为您处理轮换，并以所见即所得的方式使所有用例保持一致。

【讨论】：

LifecycleCameraController 的一个问题是缺少样板代码； API 描述页面中的内容不足以开始使用，并且在 android 应用程序中通过反复试验进行编程非常耗时……我非常感谢对此的 codelab 扩展。我会检查官方的 cameraXsample 看看是否有帮助。

我知道代码实验室有点过时了。有兴趣的可以看一下CameraX的测试app，了解控制器应该如何使用：android.googlesource.com/platform/frameworks/support/+/refs/…

感谢您的链接；但据我了解，CameraX 不喜欢干预图像数据，所以它唯一要做的就是设置 EXIF 元数据。最后我还是使用了ProcessCameraProvider类，在调用takePicture()之前设置了目标旋转，然后手动从保存的JPEG中读取EXIF数据，并相应地旋转图片以获得直上图像。即使我希望 CameraX 负责旋转图像并且不依赖 EXIF 元数据，我也会将此作为单独的答案添加。

我尝试使用 LifecycleCameraController。它没有帮助。照片仍旋转 -90 度。顺便说一句，这门课非常原始。无法设置照片/分析的分辨率，并且其他基本功能不可用。

天哪，我爱你提到 LifecycleCameraController。我删除了所有代码并用控制器替换了它。

【参考方案2】：

我也遇到了同样的情况。我用hacky方式解决了这个问题。

我的解决办法是：

    fun Bitmap.rotate(degrees: Float): Bitmap 
    val matrix = Matrix().apply  postRotate(degrees) 
    return Bitmap.createBitmap(this, 0, 0, width, height, matrix, true)
    

用法：

imageViewCapturedImage.setImageBitmap(bitmap?.rotate(90F))

【讨论】：

是的，我自己也在轮换。请参阅我对轮换助手类的回答

【参考方案3】：

我已经使用这个类来旋转图像

object CaptureImageHelper 

/**
 * This method is responsible for solving the rotation issue if exist. Also scale the images to
 * 1024x1024 resolution
 *
 * @param context       The current context
 * @param selectedImage The Image URI
 * @return Bitmap image results
 * @throws IOException
 */
@Throws(IOException::class)
fun handleSamplingAndRotationBitmap(context: Context, selectedImage: Uri?): Bitmap? 
    val MAX_HEIGHT = 1024
    val MAX_WIDTH = 1024

    // First decode with inJustDecodeBounds=true to check dimensions
    val options = BitmapFactory.Options()
    options.inJustDecodeBounds = true
    var imageStream: InputStream = context.getContentResolver().openInputStream(selectedImage!!)!!
    BitmapFactory.decodeStream(imageStream, null, options)
    imageStream.close()

    // Calculate inSampleSize
    options.inSampleSize = calculateInSampleSize(options, MAX_WIDTH, MAX_HEIGHT)

    // Decode bitmap with inSampleSize set
    options.inJustDecodeBounds = false
    imageStream = context.getContentResolver().openInputStream(selectedImage!!)!!
    var img = BitmapFactory.decodeStream(imageStream, null, options)
    img = rotateImageIfRequired(img!!, selectedImage)
    return img


/**
 * Calculate an inSampleSize for use in a [BitmapFactory.Options] object when decoding
 * bitmaps using the decode* methods from [BitmapFactory]. This implementation calculates
 * the closest inSampleSize that will result in the final decoded bitmap having a width and
 * height equal to or larger than the requested width and height. This implementation does not
 * ensure a power of 2 is returned for inSampleSize which can be faster when decoding but
 * results in a larger bitmap which isn't as useful for caching purposes.
 *
 * @param options   An options object with out* params already populated (run through a decode*
 * method with inJustDecodeBounds==true
 * @param reqWidth  The requested width of the resulting bitmap
 * @param reqHeight The requested height of the resulting bitmap
 * @return The value to be used for inSampleSize
 */
private fun calculateInSampleSize(
    options: BitmapFactory.Options,
    reqWidth: Int, reqHeight: Int
): Int 
    // Raw height and width of image
    val height = options.outHeight
    val width = options.outWidth
    var inSampleSize = 1
    if (height > reqHeight || width > reqWidth) 

        // Calculate ratios of height and width to requested height and width
        val heightRatio =
            Math.round(height.toFloat() / reqHeight.toFloat())
        val widthRatio =
            Math.round(width.toFloat() / reqWidth.toFloat())

        // Choose the smallest ratio as inSampleSize value, this will guarantee a final image
        // with both dimensions larger than or equal to the requested height and width.
        inSampleSize = if (heightRatio < widthRatio) heightRatio else widthRatio

        // This offers some additional logic in case the image has a strange
        // aspect ratio. For example, a panorama may have a much larger
        // width than height. In these cases the total pixels might still
        // end up being too large to fit comfortably in memory, so we should
        // be more aggressive with sample down the image (=larger inSampleSize).
        val totalPixels = width * height.toFloat()

        // Anything more than 2x the requested pixels we'll sample down further
        val totalReqPixelsCap = reqWidth * reqHeight * 2.toFloat()
        while (totalPixels / (inSampleSize * inSampleSize) > totalReqPixelsCap) 
            inSampleSize++
        
    
    return inSampleSize


/**
 * Rotate an image if required.
 *
 * @param img           The image bitmap
 * @param selectedImage Image URI
 * @return The resulted Bitmap after manipulation
 */
@Throws(IOException::class)
private fun rotateImageIfRequired(img: Bitmap, selectedImage: Uri): Bitmap? 
    val ei = ExifInterface(selectedImage.path)
    val orientation: Int =
        ei.getAttributeInt(ExifInterface.TAG_ORIENTATION, ExifInterface.ORIENTATION_NORMAL)
    return when (orientation) 
        ExifInterface.ORIENTATION_ROTATE_90 -> rotateImage(img, 90)
        ExifInterface.ORIENTATION_ROTATE_180 -> rotateImage(img, 180)
        ExifInterface.ORIENTATION_ROTATE_270 -> rotateImage(img, 270)
        else -> img
    


private fun rotateImage(img: Bitmap, degree: Int): Bitmap? 
    val matrix = Matrix()
    matrix.postRotate(degree.toFloat())
    val rotatedImg =
        Bitmap.createBitmap(img, 0, 0, img.width, img.height, matrix, true)
    img.recycle()
    return rotatedImg

【参考方案4】：

我也遇到了同样的问题；据我了解，从this 或this 等工单的回复来看，CameraX 背后的团队不喜欢干涉硬件返回的原始图像数据，并且非常愿意将自己限制在设置 EXIF 元数据。

所以我刚刚解决了这个问题，并从与您的代码非常相似的代码开始（嗯，很大程度上受到了 codelab 中的启发），我添加了这个：

Display d = getDisplay();
if (d != null) 
    iCapture.setTargetRotation(d.getRotation());

就在调用iCapture.takePicture() 之前（iCapture 是我的ImageCapture 用例实例）。这样可以确保 EXIF 文件元数据中的目标旋转与拍照时的实际显示旋转一致。

然后，在收到数据后（在我的情况下，在 onImageSaved() 处理程序上），我检查 EXIF 元数据的旋转，在这种情况下，手动旋转图像所需的度数并保存不同的文件以确保没有 EXIF 标记保留不连贯的值。

try 
    ExifInterface ei = new ExifInterface(tempFile.getAbsolutePath());
    if (ei.getRotationDegrees() != 0) 
        actualPicture = ImageUtil.rotateDegrees(tempFile, ei.getRotationDegrees());
    
 catch (IOException exc) 
    Log.e(TAG, "Tried to fix image rotation but could not continue: " + exc,getMessage());

其中 ImageUtil 是图像工具的自定义类，而 rotateDegrees 就是这样做的，自定义矩阵初始化如下：

//inside rotateDegrees(), degrees is the parameter to the function
Matrix m = new Matrix();
m.postRotate(degrees);

并从从原始文件导入的位图开始创建一个新位图：

Bitmap b = Bitmap.createBitmap(sourceFile, 0, 0, sourceFile.getWidth(), sourceFile.getHeight(), m, true);
b.compress(Bitmap.CompressFormat.JPEG, 85, /* a suitably-created output stream */);

不过，我希望 CameraX 直接处理图像旋转，而不依赖元数据（据他们自己承认，很少有库和工具会被读取并实际处理）。

【参考方案5】：

这个简单的代码对我有用：

Java 版本：

Context context = ... //The current Context
Camera camera = cameraProvider.bindToLifecycle(...); //The one you get after initializing the camera
ImageProxy image = ... //The one that takePicture or Analyze give you
int currentLensOrientation = ... //CameraSelector.LENS_FACING_BACK or CameraSelector.LENS_FACING_FRONT

int rotationDirection = currentLensOrientation == CameraSelector.LENS_FACING_BACK ? 1 : -1;
int constantRotation = image.getImageInfo().getRotationDegrees() - camera.getCameraInfo().getSensorRotationDegrees();
int rotationDegrees = camera.getCameraInfo().getSensorRotationDegrees() - context.getDisplay().getRotation() * 90 + constantRotation * rotationDirection;

Kotlin 版本：

val context: Context = ... //The current Context
val camera: Camera? = cameraProvider.bindToLifecycle(...) //The one you get after initializing the camera
val image: ImageProxy = ... //The one that takePicture or Analyze give you
val currentLensOrientation: Int = ... //CameraSelector.LENS_FACING_BACK or CameraSelector.LENS_FACING_FRONT

val rotationDirection = if (currentLensOrientation == CameraSelector.LENS_FACING_BACK) 1 else -1
val constantRotation = image.imageInfo.rotationDegrees - camera!!.cameraInfo.sensorRotationDegrees
val rotationDegrees = camera!!.cameraInfo.sensorRotationDegrees - context.display!!.rotation * 90 + constantRotation * rotationDirection

然后我使用rotationDegrees 来旋转CameraX 在takePicture 和analyze 回调中传递给您的ImageProxy。

如果需要，您可以在此处找到完整的 Java 代码：https://github.com/CristianDavideConte/SistemiDigitali/blob/7b40e50d8b2fbdf4e4a61edba7443da92b96c58d/app/src/main/java/com/example/sistemidigitali/views/CameraProviderView.java#L207