I am attempting to write some code that will expediently process video frames. I am receiving the frames as a System.Windows.Media.Imaging.WriteableBitmap
. For testing purposes, I am just applying a simple threshold filter that will process a BGRA format image and assign each pixel to either be black or white based on the average of the BGR pixels.
Here is my "Safe" version:
public static void ApplyFilter(WriteableBitmap Bitmap, byte Threshold)
{
// Let's just make this work for this format
if (Bitmap.Format != PixelFormats.Bgr24
&& Bitmap.Format != PixelFormats.Bgr32)
{
return;
}
// Calculate the number of bytes per pixel (should be 4 for this format).
var bytesPerPixel = (Bitmap.Format.BitsPerPixel + 7) / 8;
// Stride is bytes per pixel times the number of pixels.
// Stride is the byte width of a single rectangle row.
var stride = Bitmap.PixelWidth * bytesPerPixel;
// Create a byte array for a the entire size of bitmap.
var arraySize = stride * Bitmap.PixelHeight;
var pixelArray = new byte[arraySize];
// Copy all pixels into the array
Bitmap.CopyPixels(pixelArray, stride, 0);
// Loop through array and change pixels to black/white based on threshold
for (int i = 0; i < pixelArray.Length; i += bytesPerPixel)
{
// i=B, i+1=G, i+2=R, i+3=A
var brightness =
(byte)((pixelArray[i] + pixelArray[i+1] + pixelArray[i+2]) / 3);
var toColor = byte.MinValue; // Black
if (brightness >= Threshold)
{
toColor = byte.MaxValue; // White
}
pixelArray[i] = toColor;
pixelArray[i + 1] = toColor;
pixelArray[i + 2] = toColor;
}
Bitmap.WritePixels(
new Int32Rect(0, 0, Bitmap.PixelWidth, Bitmap.PixelHeight),
pixelArray, stride, 0
);
}
Here is what I think is a direct translation using an unsafe code block and the WriteableBitmap Back Buffer instead of the forebuffer:
public static void ApplyFilterUnsafe(WriteableBitmap Bitmap, byte Threshold)
{
// Let's just make this work for this format
if (Bitmap.Format != PixelFormats.Bgr24
&& Bitmap.Format != PixelFormats.Bgr32)
{
return;
}
var bytesPerPixel = (Bitmap.Format.BitsPerPixel + 7) / 8;
Bitmap.Lock();
unsafe
{
// Get a pointer to the back buffer.
byte* pBackBuffer = (byte*)Bitmap.BackBuffer;
for (int i = 0;
i < Bitmap.BackBufferStride*Bitmap.PixelHeight;
i+= bytesPerPixel)
{
var pCopy = pBackBuffer;
var brightness = (byte)((*pBackBuffer
+ *++pBackBuffer
+ *++pBackBuffer) / 3);
pBackBuffer++;
var toColor =
brightness >= Threshold ? byte.MaxValue : byte.MinValue;
*pCopy = toColor;
*++pCopy = toColor;
*++pCopy = toColor;
}
}
// Bitmap.AddDirtyRect(
// new Int32Rect(0,0, Bitmap.PixelWidth, Bitmap.PixelHeight));
Bitmap.Unlock();
}
This is my first foray into unsafe code blocks and pointers, so maybe the logic is not optimal.
I have tested both blocks of code on the same WriteableBitmaps using:
var threshold = Convert.ToByte(op.Result);
var copy2 = copyFrame.Clone();
Stopwatch stopWatch = new Stopwatch();
stopWatch.Start();
BinaryFilter.ApplyFilterUnsafe(copyFrame, threshold);
stopWatch.Stop();
var unsafesecs = stopWatch.ElapsedMilliseconds;
stopWatch.Reset();
stopWatch.Start();
BinaryFilter.ApplyFilter(copy2, threshold);
stopWatch.Stop();
Debug.WriteLine(string.Format("Unsafe: {1}, Safe: {0}",
stopWatch.ElapsedMilliseconds, unsafesecs));
So I am analyzing the same image. A test run of an incoming stream of video frames:
Unsafe: 110, Safe: 53
Unsafe: 136, Safe: 42
Unsafe: 106, Safe: 36
Unsafe: 95, Safe: 43
Unsafe: 98, Safe: 41
Unsafe: 88, Safe: 36
Unsafe: 129, Safe: 65
Unsafe: 100, Safe: 47
Unsafe: 112, Safe: 50
Unsafe: 91, Safe: 33
Unsafe: 118, Safe: 42
Unsafe: 103, Safe: 80
Unsafe: 104, Safe: 34
Unsafe: 101, Safe: 36
Unsafe: 154, Safe: 83
Unsafe: 134, Safe: 46
Unsafe: 113, Safe: 76
Unsafe: 117, Safe: 57
Unsafe: 90, Safe: 41
Unsafe: 156, Safe: 35
Why is my unsafe version always slower? Is it due to using the back buffer? Or am I doing something wrong?
Thanks
Maybe because your unsafe version is doing a multiply and property access:
Bitmap.BackBufferStride*Bitmap.PixelHeight
On every loop iteration. Store the result in a variable.
One further optimization, in either safe or unsafe code:
Stop dividing by 3 inside your loop. Multiply your threshold by 3 once, outside the loop. You will need to use some type other than byte
, but that shouldn't be a problem. Actually, you're already using a bigger datatype than byte
:)
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