Replace GDI+ DrawImage with PInvoked GDI and transparent PNG's

I've created an image service in C# which takes a base layer image (JPG), layers one more more transparent PNG's (32 bit), and then outputs a final JPG image. I'm trying to squeeze every last millisecond out of this function and my code is bottlenecking at the DrawImage call in GDI+. Managed code here:

// Load base image and create graphics
Image image = LoadImage(renderSettings.RenderedImageDirectory + baseLayer);

Graphics graphics = Graphics.FromImage(image);              
graphics.CompositingQuality = System.Drawing.Drawing2D.CompositingQuality.HighQuality;
graphics.InterpolationMode = System.Drawing.Drawing2D.InterpolationMode.HighQualityBicubic; 
graphics.SmoothingMode = System.Drawing.Drawing2D.SmoothingMode.HighQuality;
graphics.PixelOffsetMode = System.Drawing.Drawing2D.PixelOffsetMode.HighSpeed;

// Draw additional layers to final image
for (int i = 1; i < renderLayers.Count; i++) {
    // SLOW -- LoadImage just a utility method that returns an Image from disk or cache
    graphics.DrawImage(LoadImage(renderSettings.RenderedImageDirectory + renderLayers[i]), 0, 0, image.Width, image.Height);            
}

if (graphics != null) graphics.Dispose();

Now, I read about the performance gains obtained by calling GDI directly by P/Invoke and made an attempt at replacing the DrawImage call. I created a unit test to try to duplicate the same functionality of loading a JPG and then layering one transparent PNG on top of it.

Ref: http://social.msdn.microsoft.com/Forums/en-US/winforms/thread/29582142-0068-40dd-bd99-4b3883a76350

Bitmap sourceImage = new Bitmap("c:\\base.jpg");
Bitmap overlayImage = new Bitmap("c:\\layer1.png");

// NOTE: ImageHelper is a utility class containing all the P/Invoke stuff

// Get source image in memory
Graphics sourceImageGraphics = Graphics.FromImage(sourceImage);
IntPtr sourceImageHDC = sourceImageGraphics.GetHdc();
IntPtr sourceImageCDC = ImageHelper.CreateCompatibleDC(sourceImageHDC);
IntPtr sourceImageHandle = sourceImage.GetHbitmap();
ImageHelper.SelectObject(sourceImageCDC, sourceImageHandle);

// Get overlay image in memory
Graphics overlayImageGraphics = Graphics.FromImage(overlayImage);
IntPtr overlayImageHDC = overlayImageGraphics.GetHdc();
IntPtr overlayImageCDC = ImageHelper.CreateCompatibleDC(overlayImageHDC);
IntPtr overlayImageHandle = overlayImage.GetHbitmap();

ImageHelper.SelectObject(overlayImageCDC, overlayImageHandle);
ImageHelper.BitBlt(sourceImageHDC, 0, 0, sourceImage.Width, sourceImage.Height, overlayImageCDC, 0, 0, ImageHelper.TernaryRasterOperations.SRCAND);
ImageHelper.AlphaBlend(sourceImageHDC, 0, 0, sourceImage.Width, sourceImage.Height, overlayImageCDC, 0, 0, sourceImage.Width, sourceImage.Height, new ImageHelper.BLENDFUNCTION(ImageHelper.AC_SRC_OVER, 0, 0xff, ImageHelper.AC_SRC_ALPHA));

// Release source Image memory.
ImageHelper.DeleteDC(sourceImageCDC);
ImageHelper.DeleteObject(sourceImageHandle);
sourceImageGraphics.ReleaseHdc(sourceImageHDC);
sourceImageGraphics.Dispose();

// Release overlay Image memory.
ImageHelper.DeleteDC(overlayImageCDC);
ImageHelper.DeleteObject(overlayImageHandle);
overlayImageGraphics.ReleaseHdc(overlayImageHDC);
overlayImageGraphics.Dispose();

// Save to jpg
sourceImage.Save("c:\\output.jpg", ImageFormat.Jpeg);

But this fails to produce a layered image. Just the PNG without the base JPG. What should I be doing differently? I'm a little out of my league when in comes to straight GDI.

Solution

I ended up using SharpDX to access both the WIC and Direct2d API's. The results are impressive to say the least. When compositing with Direct2d I'm seeing increased performance as much as 400-500% over GDI+.

I also tried GDI+ and the Task Parallel Library to break up images into four quandrants and do compositing work in each core. The results weren't nearly as signficant as using SharpDX.

Here's the code I ended up using. The reference to "renderSettings" is just a configuration object. Substitute as needed along with the renderLayer image list.

/* SharpDX */
using SharpDX;
using SharpDX.Direct2D1;
using SharpDX.DirectWrite;
using SharpDX.DXGI;
using SharpDX.IO;
using SharpDX.WIC;
using AlphaMode = SharpDX.Direct2D1.AlphaMode;
using WicBitmap = SharpDX.WIC.Bitmap;
using D2DPixelFormat = SharpDX.Direct2D1.PixelFormat;
using WicPixelFormat = SharpDX.WIC.PixelFormat;
using Rectangle = System.Drawing.Rectangle;
using Bitmap = System.Drawing.Bitmap;

public Image FlattenImageDirect2d()
{

    List<string> renderLayers = new List<string>()
    {
        "image1.jpg", "image1.png", "image2.png", "image3.png", "image4.png", "image5.png", "image6.png", "image7.png"
    };

    // Base image
    string baseLayer = renderLayers[0];

    // Create WIC and D2D factories
    var wicFactory = new ImagingFactory();
    var ddFactory = new SharpDX.Direct2D1.Factory();

    // Get image size using WIC
    int baseWidth, baseHeight;
    using (var wicStream = new WICStream(wicFactory, renderDirectory + baseLayer, NativeFileAccess.Read)) {
        var jpegDecoder = new JpegBitmapDecoder(wicFactory);
        jpegDecoder.Initialize(wicStream, DecodeOptions.CacheOnDemand);
        var frame = jpegDecoder.GetFrame(0);
        baseWidth = frame.Size.Width;
        baseHeight = frame.Size.Height;
        frame.Dispose();
        jpegDecoder.Dispose();
    }

    // Resize image?
    bool resizeImage = (baseWidth != renderSettings.RenderWidth) || (baseHeight != renderSettings.RenderHeight);

    // Bitmaps and render target settings
    var wicBitmap = new WicBitmap(wicFactory, renderSettings.RenderWidth, renderSettings.RenderHeight, SharpDX.WIC.PixelFormat.Format32bppBGR, BitmapCreateCacheOption.CacheOnLoad);
    var renderTargetProperties = new RenderTargetProperties(RenderTargetType.Default, new D2DPixelFormat(Format.Unknown, AlphaMode.Unknown), 0, 0, RenderTargetUsage.None, FeatureLevel.Level_DEFAULT);
    var wicRenderTarget = new WicRenderTarget(ddFactory, wicBitmap, renderTargetProperties);

    // Create bitmap render target used to draw all images to
    SharpDX.Direct2D1.BitmapRenderTarget bitmapRenderTarget = new SharpDX.Direct2D1.BitmapRenderTarget(wicRenderTarget, CompatibleRenderTargetOptions.None, new D2DPixelFormat(Format.Unknown, AlphaMode.Premultiplied));

    // Draw render layers
    for (int i = 0; i < renderLayers.Count; i++) {
        // First layer is always a jpeg, all other subsequent layers are png's
        ImageFormat imageFormat = (i == 0) ? ImageFormat.Jpeg : ImageFormat.Png;

        using (SharpDX.WIC.BitmapSource bitmapSource = LoadWicBitmap(wicFactory, renderDirectory + renderLayers[i], imageFormat, resizeImage, renderSettings.RenderWidth, renderSettings.RenderHeight)) {

            // Convert WIC pixel format to D2D1 format
            var formatConverter = new FormatConverter(wicFactory);
            formatConverter.Initialize(bitmapSource, SharpDX.WIC.PixelFormat.Format32bppPBGRA, BitmapDitherType.None, null, 0f, BitmapPaletteType.MedianCut);

            // Create direct 2d bitmap from wic bitmap
            SharpDX.Direct2D1.Bitmap direct2DBitmap = SharpDX.Direct2D1.Bitmap.FromWicBitmap(bitmapRenderTarget, formatConverter);

            // Draw direct2d image to bitmap render target
            wicRenderTarget.BeginDraw();
            wicRenderTarget.DrawBitmap(direct2DBitmap, 1.0f, SharpDX.Direct2D1.BitmapInterpolationMode.Linear);
            wicRenderTarget.EndDraw();

            // Clean up
            formatConverter.Dispose();
            direct2DBitmap.Dispose();
        }
    }

    // Final image data
    byte[] imageData;

    // Create streams to write output to. 
    using (var memoryStream = new MemoryStream()) {
        using (var wicStream = new WICStream(wicFactory, memoryStream)) {

            // Encode wic bitmap
            var encoder = new JpegBitmapEncoder(wicFactory);
            encoder.Initialize(wicStream);

            var frameEncoder = new BitmapFrameEncode(encoder);
            frameEncoder.Initialize();
            frameEncoder.SetSize(renderSettings.RenderWidth, renderSettings.RenderHeight);
            frameEncoder.PixelFormat = WicPixelFormat.FormatDontCare;
            frameEncoder.WriteSource(wicBitmap);
            frameEncoder.Commit();
            encoder.Commit();

            // Set image data
            memoryStream.Position = 0;
            imageData = memoryStream.ToArray();

            // Clean up 
            frameEncoder.Dispose();
            encoder.Dispose();
            wicBitmap.Dispose();
            wicRenderTarget.Dispose();
            bitmapRenderTarget.Dispose();
            ddFactory.Dispose();
            wicFactory.Dispose();
            frameEncoder = null;
            encoder = null;
            wicBitmap = null;
            wicRenderTarget = null;
            bitmapRenderTarget = null;                  
            ddFactory = null;
            wicFactory = null;
        }
    }

    return Image.FromStream(new MemoryStream(imageData));
}

private BitmapSource LoadWicBitmap(ImagingFactory wicFactory, string path, ImageFormat imageFormat, bool resize, int resizeWidth = 0, int resizeHeight = 0)
{
    PngBitmapDecoder pngDecoder;
    JpegBitmapDecoder jpegDecoder;
    BitmapFrameDecode bitmapFrameDecode;

    var stream = new WICStream(wicFactory, path, NativeFileAccess.Read);

    // Load the appropriate decoder
    if (imageFormat == ImageFormat.Jpeg) {
        jpegDecoder = new JpegBitmapDecoder(wicFactory);
        jpegDecoder.Initialize(stream, DecodeOptions.CacheOnLoad);
        bitmapFrameDecode = jpegDecoder.GetFrame(0);
        jpegDecoder.Dispose();
    }
    else {
        pngDecoder = new PngBitmapDecoder(wicFactory);
        pngDecoder.Initialize(stream, DecodeOptions.CacheOnDemand);
        bitmapFrameDecode = pngDecoder.GetFrame(0);
        pngDecoder.Dispose();
    }

    // Clean up
    stream.Dispose();

    // Resize if necessary
    if (resize) {
        // Prepare scaler
        var scaler = new BitmapScaler(wicFactory);
        scaler.Initialize(bitmapFrameDecode, resizeWidth, resizeHeight, SharpDX.WIC.BitmapInterpolationMode.Fant);
        return (BitmapSource)scaler;
    }

    return (BitmapSource)bitmapFrameDecode;
}