Is there a way to easily extract the DCT coefficients (and quantization parameters) from encoded images and video? Any decoder software must be using them to decode block-DCT encoded images and video. So I'm pretty sure the decoder knows what they are. Is there a way to expose them to whomever is using the decoder?

I'm implementing some video quality assessment algorithms that work directly in the DCT domain. Currently, the majority of my code uses OpenCV, so it would be great if anyone knows of a solution using that framework. I don't mind using other libraries (perhaps libjpeg, but that seems to be for still images only), but my primary concern is to do as little format-specific work as possible (I don't want to reinvent the wheel and write my own decoders). I want to be able to open any video/image (H.264, MPEG, JPEG, etc) that OpenCV can open, and if it's block DCT-encoded, to get the DCT coefficients.

In the worst case, I know that I can write up my own block DCT code, run the decompressed frames/images through it and then I'd be back in the DCT domain. That's hardly an elegant solution, and I hope I can do better.

Presently, I use the fairly common OpenCV boilerplate to open images:

IplImage *image = cvLoadImage(filename);
// Run quality assessment metric

The code I'm using for video is equally trivial:

CvCapture *capture = cvCaptureFromAVI(filename);    
while (cvGrabFrame(capture))
{
    IplImage *frame = cvRetrieveFrame(capture);
    // Run quality assessment metric on frame
}
cvReleaseCapture(&capture);

In both cases, I get a 3-channel IplImage in BGR format. Is there any way I can get the DCT coefficients as well?