如何在不先读取整个图像的情况下原地缩放流式位图？

Question

我有一个非常图像密集的Android应用程序。我目前使用Bitmap.createScaledBitmap()来将图像缩放到所需大小。然而，这种方法要求我已经在内存中拥有原始位图，这可能相当大。

如何在不首先将整个位图写入本地内存或文件系统的情况下缩放正在下载的位图？

- emmby

---

可能是在Android上将大位图文件调整大小为缩放输出文件的重复问题。 - jww

Answer 1

这种方法会读取图像的头信息以确定其大小，然后在原地读取图像并将其缩放到所需大小，而不为完整尺寸的图像分配内存。

它还使用了BitmapFactory.Options.inPurgeable，这似乎是一个文档稀少但令人渴望的选项，可防止在使用大量位图时出现OoM异常。更新：不再使用inPurgeable，请参见Romain的此说明

它通过使用BufferedInputStream来读取图像的头信息，然后通过InputStream读取整个图像来实现。

/**
 * Read the image from the stream and create a bitmap scaled to the desired
 * size.  Resulting bitmap will be at least as large as the 
 * desired minimum specified dimensions and will keep the image proportions 
 * correct during scaling.
 */
protected Bitmap createScaledBitmapFromStream( InputStream s, int minimumDesiredBitmapWith, int minimumDesiredBitmapHeight ) {

    final BufferedInputStream is = new BufferedInputStream(s, 32*1024);
    try {
        final Options decodeBitmapOptions = new Options();
        // For further memory savings, you may want to consider using this option
        // decodeBitmapOptions.inPreferredConfig = Config.RGB_565; // Uses 2-bytes instead of default 4 per pixel

        if( minimumDesiredBitmapWidth >0 && minimumDesiredBitmapHeight >0 ) {
            final Options decodeBoundsOptions = new Options();
            decodeBoundsOptions.inJustDecodeBounds = true;
            is.mark(32*1024); // 32k is probably overkill, but 8k is insufficient for some jpgs
            BitmapFactory.decodeStream(is,null,decodeBoundsOptions);
            is.reset();

            final int originalWidth = decodeBoundsOptions.outWidth;
            final int originalHeight = decodeBoundsOptions.outHeight;

            // inSampleSize prefers multiples of 2, but we prefer to prioritize memory savings
            decodeBitmapOptions.inSampleSize= Math.max(1,Math.min(originalWidth / minimumDesiredBitmapWidth, originalHeight / minimumDesiredBitmapHeight));

        }

        return BitmapFactory.decodeStream(is,null,decodeBitmapOptions);

    } catch( IOException e ) {
        throw new RuntimeException(e); // this shouldn't happen
    } finally {
        try {
            is.close();
        } catch( IOException ignored ) {}
    }

}

- emmby

---

在调用 BitmapFactory.decodeStream 之前，您可能还需要进行显式的 GC 调用，如此答案所述：http://stackoverflow.com/questions/6718855/using-caching-to-improve-scrolling-performance-in-an-android-listview-with-images/7245318#7245318 - emmby

使用FileInputStream时，使用new/close可能比mark/reset更合适。但这个想法很好。 - kellogs

为什么在 is.reset(); 上会出现 java.io.IOException: Mark has been invalidated？ - sancho21

我也收到了“标记已失效”的错误提示，不得不将标记更改为64k才能使某些图像正常工作：is.mark(64 * 1024) - Assaf S.

Answer 2

这是我基于 @emmby 的解决方案（感谢你！）的版本。 我增加了第二阶段，在该阶段中，您将缩小的位图再次缩放以完全匹配所需的尺寸。 我的版本接受文件路径而不是流。

protected Bitmap createScaledBitmap(String filePath, int desiredBitmapWith, int desiredBitmapHeight) throws IOException, FileNotFoundException {
    BufferedInputStream imageFileStream = new BufferedInputStream(new FileInputStream(filePath));
    try {
        // Phase 1: Get a reduced size image. In this part we will do a rough scale down
        int sampleSize = 1;
        if (desiredBitmapWith > 0 && desiredBitmapHeight > 0) {
            final BitmapFactory.Options decodeBoundsOptions = new BitmapFactory.Options();
            decodeBoundsOptions.inJustDecodeBounds = true;
            imageFileStream.mark(64 * 1024);
            BitmapFactory.decodeStream(imageFileStream, null, decodeBoundsOptions);
            imageFileStream.reset();
            final int originalWidth = decodeBoundsOptions.outWidth;
            final int originalHeight = decodeBoundsOptions.outHeight;
            // inSampleSize prefers multiples of 2, but we prefer to prioritize memory savings
            sampleSize = Math.max(1, Math.max(originalWidth / desiredBitmapWith, originalHeight / desiredBitmapHeight));
        }
        BitmapFactory.Options decodeBitmapOptions = new BitmapFactory.Options();
        decodeBitmapOptions.inSampleSize = sampleSize;
        decodeBitmapOptions.inPreferredConfig = Bitmap.Config.RGB_565; // Uses 2-bytes instead of default 4 per pixel

        // Get the roughly scaled-down image
        Bitmap bmp = BitmapFactory.decodeStream(imageFileStream, null, decodeBitmapOptions);

        // Phase 2: Get an exact-size image - no dimension will exceed the desired value
        float ratio = Math.min((float)desiredBitmapWith/ (float)bmp.getWidth(), (float)desiredBitmapHeight/ (float)bmp.getHeight());
        int w =(int) ((float)bmp.getWidth() * ratio);
        int h =(int) ((float)bmp.getHeight() * ratio);
        return Bitmap.createScaledBitmap(bmp, w,h, true);

    } catch (IOException e) {
        throw e;
    } finally {
        try {
            imageFileStream.close();
        } catch (IOException ignored) {
        }
    }
}

- Assaf S.