Introduction
I was busy trying to write a motion detection algorithm in C#. To determine motion, I compared change values from a previous image by calculating the absolute value of the pixel difference. For doing this, grayscale is easier and probably faster.
Background
If MIT thinks using grayscale for motion detection is a good idea, then I am definitely on the right track with my motion detection algorithm. For a reference, see the Cog project.
I got the code for the image loop from here because I originally attempted to use SetPixel
which is painfully slow.
Using the code
If you use this code in a multithreaded environment, be aware that the parameter image will be locked until this method unlocks it. This will raise an exception when another code attempts to lock the bitmap, such as a customer OnPaint
. To avoid this, you can always make a copy of the bitmap.
Here is the entire method:
public Bitmap processImage(Bitmap image){
Bitmap returnMap = new Bitmap(image.Width, image.Height,
PixelFormat.Format32bppArgb);
BitmapData bitmapData1 = image.LockBits(new Rectangle(0, 0,
image.Width, image.Height),
ImageLockMode.ReadOnly,
PixelFormat.Format32bppArgb);
BitmapData bitmapData2 = returnMap.LockBits(new Rectangle(0, 0,
returnMap.Width, returnMap.Height),
ImageLockMode.ReadOnly,
PixelFormat.Format32bppArgb);
int a = 0;
unsafe {
byte* imagePointer1 = (byte*)bitmapData1.Scan0;
byte* imagePointer2 = (byte*)bitmapData2.Scan0;
for(int i = 0; i < bitmapData1.Height; i++) {
for(int j = 0; j < bitmapData1.Width; j++) {
a = (imagePointer1[0] + imagePointer1[1] +
imagePointer1[2])/3;
imagePointer2[0] = (byte)a;
imagePointer2[1] = (byte)a;
imagePointer2[2] = (byte)a;
imagePointer2[3] = imagePointer1[3];
imagePointer1 += 4;
imagePointer2 += 4;
}
imagePointer1 += bitmapData1.Stride -
(bitmapData1.Width * 4);
imagePointer2 += bitmapData1.Stride -
(bitmapData1.Width * 4);
}
}
returnMap.UnlockBits(bitmapData2);
image.UnlockBits(bitmapData1);
return returnMap;
}
The PixelFormat
is crucial. This format determines the layout of the bitmap data. This is why the literal 4 is hard-coded (actually, it is hard-coded because I am lazy). If you use a different PixelFormat
, you will need to determine the layout and offset for those formats.
BitmapData bitmapData1 = image.LockBits(new Rectangle(0, 0,
image.Width, image.Height),
ImageLockMode.ReadOnly,
PixelFormat.Format32bppArgb);
BitmapData bitmapData2 = returnMap.LockBits(new Rectangle(0, 0,
returnMap.Width, returnMap.Height),
ImageLockMode.ReadOnly, PixelFormat.Format32bppArgb);
This code locks the bitmap and returns its image data. Since I am converting the entire image to grayscale, I lock the entire bitmap. It is possible to lock a smaller area.
byte* imagePointer1 = (byte*)bitmapData1.Scan0;
byte* imagePointer2 = (byte*)bitmapData2.Scan0;
Pointer arithmetic is unsafe in C# :(. Too bad because it is fast! Although, it is strange getting Access Violation errors in C#. If you change the code, test thoroughly using Mathematics to make sure you don't overwrite any protected memory. Scan0
is the pointer to the first byte of the bitmap's data.
for(int i = 0; i < bitmapData1.Height; i++) {
for(int j = 0; j < bitmapData1.Width; j++) {
The astute will immediately recognize this will only work if the bitmaps are the same size. And fortunately, this always is since this method will create the returnMap
bitmap based on the input bitmap. An incorrect loop will also cause an Access Violation.
imagePointer2[0] = (byte)a;
imagePointer2[1] = (byte)a;
imagePointer2[2] = (byte)a;
imagePointer2[3] = imagePointer1[3];
See the comments in the code snippet above, to understand the layout of the bitmap data.
a = (imagePointer1[0] + imagePointer1[1] + imagePointer1[2])/3;
Average the three color components in the original bitmap. Keep the alpha the same, otherwise your grayscale will also cause undesired blending.
imagePointer1 += 4;
imagePointer2 += 4;
Move forward 4 bytes:
imagePointer1 += bitmapData1.Stride - (bitmapData1.Width * 4);
imagePointer2 += bitmapData1.Stride - (bitmapData1.Width * 4);
This is definitely the most confusing part. See this for a picture of what is happening. The width of a bitmap is actually the composed width and an unused buffer to pad the bitmap to be a multiple of 4. Width + buffer = stride. It works, so good enough for me.
returnMap.UnlockBits(bitmapData2);
image.UnlockBits(bitmapData1);
return returnMap;
UnlockBits
unlocks the bitmap data, probably a good thing to do.
Points of Interest
A good place to find answers to hard problems: Google.
People much smarter than me: MIT Cog Project.
Possible Errors
I use the bitmap data from bitmapData1
for both bitmapData1
and bitmapData2
. This could lead to unexpected errors if the data were some how different. However, in testing, everything works fine. A lot more prudent coder would verify a lot of things before putting this sort of code into anything critical.
If there are any more errors, I am quite sure that the overly critical among you will gladly point them out as glaring criticisms.