c# GDI Edge Whitespace Detection Algorithm

Question

I am looking for a solution for detecting edge whitespace of c# bitmap, from the c# managed GDI+ library.

The images would be either transparent or white, most of the 400x pictures are 8000x8000px with about 2000px whitespace around the edges.

What would be the most efficient way of finding out the edges, x, y, height and width coordinates? I tried a go pixel by pixel but was finding it very slow.

Update to solution --Added left/right/top/bottom bounds

Problems with images detail center images, now crops any transparent (0%) or white (#FFFFFF) pixels.

var top = bitmap.Height;
var left = bitmap.Width;
var right = 0;
var bottom = 0;

...

var pData = pData0 + (y * data.Stride) + (x * 4);
var xyAlpha = pData[3];
var xyBlue = pData[0];
var xyGreen = pData[1];
var xyRed = pData[2];
if ((xyAlpha > 0) || (xyRed != 255 && xyGreen != 255 && xyBlue != 255)) {
    if (y < top)
        top = y;
    if (y > bottom)
        bottom = y;
    if (x < left)
        left = x;
    if (x > right)
        right = x;
}

...

var cropWidth = right - left;
var cropHeight = bottom - top;
var cropX = top;
var cropY = left;

var cacheBitmap = new Bitmap(cropWidth, cropHeight, PixelFormat.Format32bppArgb);
using (var cacheGraphics = Graphics.FromImage(cacheBitmap)) {
    cacheGraphics.DrawImage(context.Image, new Rectangle(0, 0, cropWidth, cropHeight), cropX, cropY, cropWidth, cropHeight, GraphicsUnit.Pixel);
}

Answer 1

A great GDI+ resource is Bob Powells GDI+ FAQ!

You didn't say how you accessed the pixels in the image so I will assume that you used the slow GetPixel methods. You can use pointers and LockBits to access pixels in a faster way: see Bob Powells explanation of LockBits - This will require an unsafe code block - if you don't want this or you do not have FullTrust you can use the trick explained here: Pointerless Image Processing in .NET by J. Dunlap

The below code uses the LockBits approach (for the PixelFormat.Format32bppArgb) and will fill the start and end Points with the value where the first and last pixels in an image are discovered that do not have the color described in the argument color. The method also ignores completely transparent pixels which is useful if you want to detect the area of an image where the visible 'content' starts.

    Point start = Point.Empty;
    Point end = Point.Empty;

    int bitmapWidth = bmp.Width;
    int bitmapHeight = bmp.Height;

    #region find start and end point
    BitmapData data = bmp.LockBits(new Rectangle(0, 0, bitmapWidth, bitmapHeight), ImageLockMode.ReadOnly, PixelFormat.Format32bppArgb);
    try
    {
        unsafe
        {
            byte* pData0 = (byte*)data.Scan0;
            for (int y = 0; y < bitmapHeight; y++)
            {
                for (int x = 0; x < bitmapWidth; x++)
                {
                    byte* pData = pData0 + (y * data.Stride) + (x * 4);

                    byte xyBlue = pData[0];
                    byte xyGreen = pData[1];
                    byte xyRed = pData[2];
                    byte xyAlpha = pData[3];


                    if (color.A != xyAlpha
                            || color.B != xyBlue
                            || color.R != xyRed
                            || color.G != xyGreen)
                    {
                        //ignore transparent pixels
                        if (xyAlpha == 0)
                            continue;
                        if (start.IsEmpty)
                        {
                            start = new Point(x, y);
                        }
                        else if (start.Y > y)
                        {
                            start.Y = y;
                        }
                        if (end.IsEmpty)
                        {
                            end = new Point(x, y);
                        }
                        else if (end.X < x)
                        {
                            end.X = x;
                        }
                        else if (end.Y < y)
                        {
                            end.Y = y;
                        }
                    }
                }
            }
        }
    }
    finally
    {
        bmp.UnlockBits(data);
    }
    #endregion

Answer 2

I'd first make sure to use the LockBits method described by Patrick. Second I'd check the pixels on the middle lines to quickly determine the edges. By middle lines I mean, if you have say for example a 2000x1000 image, you'd look first along horizontal line number 500 (out of 1000) to find the left and right limits, then along vertical line number 1000 (out of 2000) to find the top and bottom limits. It should be very fast this way.