C# - Converting 8-bit or 16-bit grayscale raw pixel data

Question

I need to be able to convert 8-bit or 16-bit grayscale pixel data into a file format that the .NET framework can support.

The data I have available is the width, height, orientation (bottom-left) and the pixel format as 4096 shades of gray (12-bit resolution) packed in 2 bytes per pixel.

So for example each pixel ranges from 0 to 4096, and each pixel is 2 bytes.

I have already tried using PixelFormat.Format16bppGrayScale with the Bitmap constructor, and it throws a GDI+ exception. Everything I have read says that this format is not supported and that MSDN is wrong.

I want to convert this pixel buffer into a .NET Bitmap format (such as Format32bppArgb) with as little image quality loss as possible.

Anyone know how?

Solution 1

See the example below, which precomputes a lookup table (LUT) and uses that to convert each pixel. This version covers your 12-bit case; for 8-bit the code is very similar, but it is difficult to generalize across pixel formats.

A conversion from 12-bit GS to effectively 8-bit GS will lose data. However, you can adjust the LUT table to focus on a smaller range of input values with better contrast (ex. DICOM Window Center/Window Width).

class Program
{
    static void Main( string[] args )
    {
        // Test driver - create a Wedge, convert to Bitmap, save to file
        //
        int width = 4095;
        int height = 1200;
        int bits = 12;

        byte[] wedge = Wedge( width, height, bits );

        Bitmap bmp = Convert( wedge, width, height, bits );

        string file = "wedge.png";

        bmp.Save( file );

        Process.Start( file );
    }

    static Bitmap Convert( byte[] input, int width, int height, int bits )
    {
        // Convert byte buffer (2 bytes per pixel) to 32-bit ARGB bitmap

        var bitmap = new Bitmap( width, height, PixelFormat.Format32bppArgb );

        var rect = new Rectangle( 0, 0, width, height );

        var lut = CreateLut( bits );

        var bitmap_data = bitmap.LockBits( rect, ImageLockMode.WriteOnly, bitmap.PixelFormat );

        ConvertCore( width, height, bits, input, bitmap_data, lut );

        bitmap.UnlockBits( bitmap_data );

        return bitmap;
    }

    static unsafe void ConvertCore( int width, int height, int bits, byte[] input, BitmapData output, uint[] lut )
    {
        // Copy pixels from input to output, applying LUT

        ushort mask = (ushort)( ( 1 << bits ) - 1 );

        int in_stride = output.Stride;
        int out_stride = width * 2;

        byte* out_data = (byte*)output.Scan0;

        fixed ( byte* in_data = input )
        {
            for ( int y = 0; y < height; y++ )
            {
                uint* out_row = (uint*)( out_data + ( y * in_stride ) );

                ushort* in_row = (ushort*)( in_data + ( y * out_stride ) );

                for ( int x = 0; x < width; x++ )
                {
                    ushort in_pixel = (ushort)( in_row[ x ] & mask );

                    out_row[ x ] = lut[ in_pixel ];
                }
            }
        }
    }

    static uint[] CreateLut( int bits )
    {
        // Create a linear LUT to convert from grayscale to ARGB

        int max_input = 1 << bits;

        uint[] lut = new uint[ max_input ];

        for ( int i = 0; i < max_input; i++ )
        {
            // map input value to 8-bit range
            //
            byte intensity = (byte)( ( i * 0xFF ) / max_input );

            // create ARGB output value A=255, R=G=B=intensity
            //
            lut[ i ] = (uint)( 0xFF000000L | ( intensity * 0x00010101L ) );
        }

        return lut;
    }

    static byte[] Wedge( int width, int height, int bits )
    {
        // horizontal wedge

        int max = 1 << bits;

        byte[] pixels = new byte[ width * height * 2 ];

        for ( int y = 0; y < height; y++ )
        {
            for ( int x = 0; x < width; x++ )
            {
                int pixel = x % max;

                int addr = ( ( y * width ) + x ) * 2;

                pixels[ addr + 1 ] = (byte)( ( pixel & 0xFF00 ) >> 8 );
                pixels[ addr + 0 ] = (byte)( ( pixel & 0x00FF ) );
            }
        }

        return pixels;
    }
}

Solution 2

Spoof a 16b format & use ColorMatrix to map it correctly before display.

I haven't done performance tests of this approach on Windows, but on other platforms (e.g., Android) where I needed efficient memory storage and rapid remapping of different ranges in the 12b or 16b data I've made good use of this technique.

I tell it my 12/16b grayscale data is really RGB565, so that it's happy serializing, deserializing & other manipulations. When I need to display I pass it through a ColorMatrix which maps the appropriate window to an 8b grayscale in ARGB8888.

If anyone wants to try this I'll post my mapping algorithm.

Comments

[Trevor Elliott]

I can't copy the pixel data verbatim. The Bitmap constructor requires a PixelFormat to be specified and the source pixel format is not supported. I need to convert the actual pixel color information from 12-bit grayscale to 24-bit or 32-bit RGB. Ideally with no quality loss.

[James Johnston]

Then use LockBits; it should help you do exactly what you need. You will lose quality, however: both pixel formats you want to convert to only have 8 bits per channel. So you'll lose 4 of your 12 bits per pixel. You may have a lot of difficulty retaining full quality: msdn.microsoft.com/en-us/library/… says "PixelFormat48bppRGB, [...] use 16 bits per ... channel. GDI+ version 1.0 and 1.1 can read 16-bits-per-channel images, but such images are converted to an 8-bits-per-channel format for processing, displaying, and saving."

[Trevor Elliott]

I am already trying to use LockBits. LockBits does not do conversion. I need to write an algorithm to convert say 8-bit grayscale bytes to 24-bit RGB bytes.

[James Johnston]

LockBits returns a BitmapData object. BitmapData.Scan0 is a pointer to the first row in the bitmap. You can do the conversion yourself; if the pixel format is 24-bit RGB then just set each of the 3 bytes to the same grayscale value from the input bitmap.