I am trying to create a TImage with alpha transparency using code. In this example, an anti-aliased circle.
I create an 8bit opacity map for the circle and then apply it to the TImage's TBitmap using this piece of code:
type
TOpacityMap = Array[0..4095] of PByteArray;
procedure DrawAntiAliasedCircle(srcBitmap: TBitmap; FillColor : TAlphaColor; CenterX, CenterY, Radius, LineWidth, Feather: single);
var
FillR : Integer;
FillG : Integer;
FillB : Integer;
FillRGB : Integer;
OpacityMap : TOpacityMap;
AlphaScanLine : Array[0..4095] of TAlphaColor;
bitmapData : FMX.Graphics.TBitmapData;
tmpScanLine : Pointer;
X,Y : Integer;
tmpMS : TMemoryStream;
begin
{Initialization}
FillR := TAlphaColorRec(FillColor).R;
FillG := TAlphaColorRec(FillColor).G;
FillB := TAlphaColorRec(FillColor).B;
CreateAntiAliasedCircleOpacityMap(OpacityMap, srcBitmap.Width, srcBitmap.Height, CenterX, CenterY, Radius, LineWidth, Feather);
{create image based on opacity map and free memory}
If srcBitmap.Map(TMapAccess.Write, bitmapData) then
try
FillRGB := (FillR shl 16)+(FillG shl 8)+FillB;
for Y := 0 to srcBitmap.Height-1 do
begin
for X := 0 to srcBitmap.Width-1 do
AlphaScanLine[X] := (OpacityMap[Y][X] shl 24)+FillRGB; // Opacity
tmpScanLine := bitmapData.GetScanline(Y);
AlphaColorToScanLine(@AlphaScanLine,tmpScanLine,srcBitmap.Width,srcBitmap.PixelFormat);
FreeMem(OpacityMap[Y]);
end;
finally
srcBitmap.Unmap(bitmapData);
end;
// Work-around fix
{tmpMS := TMemoryStream.Create;
srcBitmap.SaveToStream(tmpMS);
srcBitmap.LoadFromStream(tmpMS);
tmpMS.Free;}
end;
The result is the image on the left. The actual TBitmap seems to be good, calling "srcBitmap.SaveToFile('circle.png')" results in a PNG file with a valid alpha channel.
I can work-around this issue by simply saving/loading the bitmap using a TMemoryStream.
How do I get the desired image on the right without the performance penalty of passing the image through a TMemoryStream?
These screenshots are from the minimal example project demonstrating this issue :
https://github.com/bLightZP/AntiAliasedCircle
edit #1 :
The github code linked above has been updated with an optimized (about 25% faster) version of Tom Brunberg's suggested fix.