There are plenty of places (including here) to show how to use Renderscript to blur images, as such:
@TargetApi(VERSION_CODES.JELLY_BEAN_MR1)
public static Bitmap renderScriptBlur(Context context, Bitmap srcBitmap, @FloatRange(from = 0.0f, to = 25.0f) float radius) {
if (srcBitmap == null)
return null;
Bitmap outputBitmap = null;
RenderScript rs = null;
try {
rs = RenderScript.create(context);
outputBitmap = Bitmap.createBitmap(srcBitmap.getWidth(), srcBitmap.getHeight(), Bitmap.Config.ARGB_8888);
final Canvas canvas = new Canvas(outputBitmap);
canvas.drawBitmap(srcBitmap, 0, 0, null);
Allocation overlayAlloc = Allocation.createFromBitmap(rs, outputBitmap);
ScriptIntrinsicBlur blur = ScriptIntrinsicBlur.create(rs, Element.U8_4(rs));
blur.setInput(overlayAlloc);
blur.setRadius(radius);
blur.forEach(overlayAlloc);
overlayAlloc.copyTo(outputBitmap);
return outputBitmap;
} catch (Exception ex) {
if (outputBitmap != null)
outputBitmap.recycle();
return srcBitmap;
} finally {
if (rs != null)
rs.destroy();
}
}
Usually it works great, but when using some images and/or radius-settings, the output image has artifacts that look like scan-lines:
I've found that there is a nicer blurring solutions (like here), but they don't use Renderscript and are also slow and memory-consuming.
I've also tried to make the input image smaller, but the output still has scanlines artifacts.
Finally, I've also reported about this, here.
Is it possible to use Renderscript to blur images without those Artifcats? Is there anything wrong with what I wrote?
RenderScript framework is basically based on data parallel computation. It distributes your application workload on all the processors available on your device like multi-core CPUs or GPUs.
The problem was with the algorithm I used. Thanks to this github project, I've found the issue (which is probably not using the correct type fot the allocation) and used a nicer approach:
private static final AtomicReference<RenderScript> sRenderscript = new AtomicReference<>();
public static Bitmap blur(Context context, Bitmap bitmap) {
return blur(context, bitmap, 4, false, false);
}
public static Bitmap blur(Context context, Bitmap bitmap, float radius) {
return blur(context, bitmap, radius, false, false);
}
public static Bitmap blur(Context context, Bitmap bitmapOriginal, @FloatRange(from = 0.0f, to = 25.0f) float radius, boolean overrideOriginal, boolean recycleOriginal) {
if (bitmapOriginal == null || bitmapOriginal.isRecycled())
return null;
RenderScript rs = sRenderscript.get();
if (rs == null)
if (!sRenderscript.compareAndSet(null, rs = RenderScript.create(context)) && rs != null)
rs.destroy();
else
rs = sRenderscript.get();
final Bitmap inputBitmap = bitmapOriginal.getConfig() == Config.ARGB_8888 ? bitmapOriginal : bitmapOriginal.copy(Config.ARGB_8888, true);
final Bitmap outputBitmap = overrideOriginal ? bitmapOriginal : Bitmap.createBitmap(bitmapOriginal.getWidth(), bitmapOriginal.getHeight(), Config.ARGB_8888);
final Allocation input = Allocation.createFromBitmap(rs, inputBitmap);
final Allocation output = Allocation.createTyped(rs, input.getType());
final ScriptIntrinsicBlur script = ScriptIntrinsicBlur.create(rs, Element.U8_4(rs));
script.setRadius(radius);
script.setInput(input);
script.forEach(output);
if (recycleOriginal && !overrideOriginal)
bitmapOriginal.recycle();
output.copyTo(outputBitmap);
return outputBitmap;
}
Now it all works nicely.
The artifact in the original version is because the same input Allocation was used as the output Allocation for IntrinsicBlur:
blur.setInput(overlayAlloc);
blur.setRadius(radius);
blur.forEach(overlayAlloc);
forEach(aOut) computes the Gaussian blur and saves the result to the output Allocation. Since the algorithm requires information about the neighbors, doing blur in place may corrupt the input data for sub-sequent calculations.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With