Informikon

I've mentioned that in my code for decoding i-frames from an mpeg stream, I used a buffer for the decoded yuv data that I passed to the constructor of a Bitmap object. If you look at the possible constructors for the Bitmap object, none can take raw yuv data. You have to perform a colorspace tranform but then if you save the Bitmap object as jpeg, the save method has to perform the reverse operation. Since I just wanted to look at the result, I tried to created a Direct3D surface with a fourcc code as a pixel format but my hardware wouldn't allow it (an integrate vidcard on a laptop). So I went back to the antiques store and used a clipper object for a DirectDraw surface to do the job. Here's the code for the InitializeGraphics methods which is passed a pointer to the raw yuv data as argument.

// we need this to create surfaces
SurfaceDescription description = new SurfaceDescription();

// our DirectDraw device
device = new Device();

// set the cooperative level to normal (ie not fullscreen)
device.SetCooperativeLevel(this, CooperativeLevelFlags.Normal);

// set the display mode width and height, and 32 bit color depth.
// you might want to change this to your monitor settings
device.SetDisplayMode(800, 600, 32, 0, false);

// create a primary surface
description.SurfaceCaps.PrimarySurface = true;

// create the primary surface
primary = new Surface(description, device);

// set a clipper object around our form
Clipper clipper = new Clipper(device);
clipper.Window = this;
primary.Clipper = clipper;

// clear out the SurfaceDescription structure.
description.Clear();

// our similated backBuffer will cover the client area
description.Width = this.ClientRectangle.Width;
description.Height = this.ClientRectangle.Height;

// create a fourcc code for NV12 and set the surf descr
int NV12 = MakeFourCC('N', 'V', '1', '2' );
PixelFormat pf = new PixelFormat();
pf.FourCC = NV12;
pf.FourCcIsValid = true;
description.PixelFormatStructure = pf;

// we want an offscreen surface in vid mem
SurfaceCaps caps = new SurfaceCaps();
caps.OffScreenPlain = true;
caps.VideoMemory = true;
description.SurfaceCaps = caps;

// Create the backbuffer
backBuffer = new Surface(description, device);

---

Bookmark it...

I have been doing some low-level video processing recently. I thought that since the I-frames of an mpeg movie are very similar to a jpeg image, it would be simple to extract these. While using a SampleGrabber in DirectShow or, I suppose, some switches with ffmpeg, it's probably not too difficult, I wanted a more "basic" approach.

So I started to study the Mpeg-1 standard carefully and write some code in C#. After awhile, my code parses the headers (pack, system, and packet) of a system stream, finds its way to an I-frame, decodes the DCT coefficients, feeds it to the IDCT, and stuffs the result in a buffer that I pass to a constructor of Bitmap object that I save.

I'm not a big fan (that's an understatement) of software patents and since the Mpeg standards are replete with these, I was happy with an abstract view like a DirectShow filter, for example. But using FFMeg has forced me to pay more attention to low-level stuff.

It seems that most Mpeg codecs are based on the "reference codec", I didnt' really look at it. I just read the standard and wrote code to do the task. For example, the code for the IDCT is a direct translation of the mathematical formula, hence the result is not very efficient. But it was certainly a good learning exercise.

---

Bookmark it...