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?
Well, I did a bit of reading and my original question seems to be an instance of wishful thinking.
Basically, it's not possible to get the DCT coefficients from H.264 video frames for the simple reason that H.264 doesn't use DCT. It uses a different transform (integer transform). Next, the coefficients for that transform don't necessarily change on a frame-by-frame basis -- H.264 is smarter cause it splits up frames into slices. It should be possible to get those coefficients through a special decoder, but I doubt OpenCV exposes it for the user.
For JPEG, things are a bit more positive. As I suspected, libjpeg exposes the DCT coefficients for you. I wrote a small app to show that it works (source at the end). It makes a new image using the DC term from each block. Because the DC term is equal to the block average (after proper scaling), the DC images are downsampled versions of the input JPEG image.
EDIT: fixed scaling in source
Original image (512 x 512):
DC images (64x64): luma Cr Cb RGB
Source (C++):
#include <stdio.h>
#include <assert.h>
#include <cv.h>
#include <highgui.h>
extern "C"
{
#include "jpeglib.h"
#include <setjmp.h>
}
#define DEBUG 0
#define OUTPUT_IMAGES 1
/*
* Extract the DC terms from the specified component.
*/
IplImage *
extract_dc(j_decompress_ptr cinfo, jvirt_barray_ptr *coeffs, int ci)
{
jpeg_component_info *ci_ptr = &cinfo->comp_info[ci];
CvSize size = cvSize(ci_ptr->width_in_blocks, ci_ptr->height_in_blocks);
IplImage *dc = cvCreateImage(size, IPL_DEPTH_8U, 1);
assert(dc != NULL);
JQUANT_TBL *tbl = ci_ptr->quant_table;
UINT16 dc_quant = tbl->quantval[0];
#if DEBUG
printf("DCT method: %x\n", cinfo->dct_method);
printf
(
"component: %d (%d x %d blocks) sampling: (%d x %d)\n",
ci,
ci_ptr->width_in_blocks,
ci_ptr->height_in_blocks,
ci_ptr->h_samp_factor,
ci_ptr->v_samp_factor
);
printf("quantization table: %d\n", ci);
for (int i = 0; i < DCTSIZE2; ++i)
{
printf("% 4d ", (int)(tbl->quantval[i]));
if ((i + 1) % 8 == 0)
printf("\n");
}
printf("raw DC coefficients:\n");
#endif
JBLOCKARRAY buf =
(cinfo->mem->access_virt_barray)
(
(j_common_ptr)cinfo,
coeffs[ci],
0,
ci_ptr->v_samp_factor,
FALSE
);
for (int sf = 0; (JDIMENSION)sf < ci_ptr->height_in_blocks; ++sf)
{
for (JDIMENSION b = 0; b < ci_ptr->width_in_blocks; ++b)
{
int intensity = 0;
intensity = buf[sf][b][0]*dc_quant/DCTSIZE + 128;
intensity = MAX(0, intensity);
intensity = MIN(255, intensity);
cvSet2D(dc, sf, (int)b, cvScalar(intensity));
#if DEBUG
printf("% 2d ", buf[sf][b][0]);
#endif
}
#if DEBUG
printf("\n");
#endif
}
return dc;
}
IplImage *upscale_chroma(IplImage *quarter, CvSize full_size)
{
IplImage *full = cvCreateImage(full_size, IPL_DEPTH_8U, 1);
cvResize(quarter, full, CV_INTER_NN);
return full;
}
GLOBAL(int)
read_JPEG_file (char * filename, IplImage **dc)
{
/* This struct contains the JPEG decompression parameters and pointers to
* working space (which is allocated as needed by the JPEG library).
*/
struct jpeg_decompress_struct cinfo;
struct jpeg_error_mgr jerr;
/* More stuff */
FILE * infile; /* source file */
/* In this example we want to open the input file before doing anything else,
* so that the setjmp() error recovery below can assume the file is open.
* VERY IMPORTANT: use "b" option to fopen() if you are on a machine that
* requires it in order to read binary files.
*/
if ((infile = fopen(filename, "rb")) == NULL) {
fprintf(stderr, "can't open %s\n", filename);
return 0;
}
/* Step 1: allocate and initialize JPEG decompression object */
cinfo.err = jpeg_std_error(&jerr);
/* Now we can initialize the JPEG decompression object. */
jpeg_create_decompress(&cinfo);
/* Step 2: specify data source (eg, a file) */
jpeg_stdio_src(&cinfo, infile);
/* Step 3: read file parameters with jpeg_read_header() */
(void) jpeg_read_header(&cinfo, TRUE);
/* We can ignore the return value from jpeg_read_header since
* (a) suspension is not possible with the stdio data source, and
* (b) we passed TRUE to reject a tables-only JPEG file as an error.
* See libjpeg.txt for more info.
*/
/* Step 4: set parameters for decompression */
/* In this example, we don't need to change any of the defaults set by
* jpeg_read_header(), so we do nothing here.
*/
jvirt_barray_ptr *coeffs = jpeg_read_coefficients(&cinfo);
IplImage *y = extract_dc(&cinfo, coeffs, 0);
IplImage *cb_q = extract_dc(&cinfo, coeffs, 1);
IplImage *cr_q = extract_dc(&cinfo, coeffs, 2);
IplImage *cb = upscale_chroma(cb_q, cvGetSize(y));
IplImage *cr = upscale_chroma(cr_q, cvGetSize(y));
cvReleaseImage(&cb_q);
cvReleaseImage(&cr_q);
#if OUTPUT_IMAGES
cvSaveImage("y.png", y);
cvSaveImage("cb.png", cb);
cvSaveImage("cr.png", cr);
#endif
*dc = cvCreateImage(cvGetSize(y), IPL_DEPTH_8U, 3);
assert(dc != NULL);
cvMerge(y, cr, cb, NULL, *dc);
cvReleaseImage(&y);
cvReleaseImage(&cb);
cvReleaseImage(&cr);
/* Step 7: Finish decompression */
(void) jpeg_finish_decompress(&cinfo);
/* We can ignore the return value since suspension is not possible
* with the stdio data source.
*/
/* Step 8: Release JPEG decompression object */
/* This is an important step since it will release a good deal of memory. */
jpeg_destroy_decompress(&cinfo);
fclose(infile);
return 1;
}
int
main(int argc, char **argv)
{
int ret = 0;
if (argc != 2)
{
fprintf(stderr, "usage: %s filename.jpg\n", argv[0]);
return 1;
}
IplImage *dc = NULL;
ret = read_JPEG_file(argv[1], &dc);
assert(dc != NULL);
IplImage *rgb = cvCreateImage(cvGetSize(dc), IPL_DEPTH_8U, 3);
cvCvtColor(dc, rgb, CV_YCrCb2RGB);
#if OUTPUT_IMAGES
cvSaveImage("rgb.png", rgb);
#else
cvNamedWindow("DC", CV_WINDOW_AUTOSIZE);
cvShowImage("DC", rgb);
cvWaitKey(0);
#endif
cvReleaseImage(&dc);
cvReleaseImage(&rgb);
return 0;
}
You can use, libjpeg to extract dct data of your jpeg file, but for h.264 video file, I can't find any open source code that give you dct data (actully Integer dct data). But you can use h.264 open source software like JM, JSVM or x264. In these two source file, you have to find their specific function that make use of dct function, and change it to your desire form, to get your output dct data.
For Image:
use the following code, and after read_jpeg_file( infilename, v, quant_tbl ), v and quant_tbl will have dct data and quantization table of your jpeg image respectively.
I used Qvector to store my output data, change it to your preferred c++ array list.
#include <iostream>
#include <stdio.h>
#include <jpeglib.h>
#include <stdlib.h>
#include <setjmp.h>
#include <fstream>
#include <QVector>
int read_jpeg_file( char *filename, QVector<QVector<int> > &dct_coeff, QVector<unsigned short> &quant_tbl)
{
struct jpeg_decompress_struct cinfo;
struct jpeg_error_mgr jerr;
FILE * infile;
if ((infile = fopen(filename, "rb")) == NULL) {
fprintf(stderr, "can't open %s\n", filename);
return 0;
}
cinfo.err = jpeg_std_error(&jerr);
jpeg_create_decompress(&cinfo);
jpeg_stdio_src(&cinfo, infile);
(void) jpeg_read_header(&cinfo, TRUE);
jvirt_barray_ptr *coeffs_array = jpeg_read_coefficients(&cinfo);
for (int ci = 0; ci < 1; ci++)
{
JBLOCKARRAY buffer_one;
JCOEFPTR blockptr_one;
jpeg_component_info* compptr_one;
compptr_one = cinfo.comp_info + ci;
for (int by = 0; by < compptr_one->height_in_blocks; by++)
{
buffer_one = (cinfo.mem->access_virt_barray)((j_common_ptr)&cinfo, coeffs_array[ci], by, (JDIMENSION)1, FALSE);
for (int bx = 0; bx < compptr_one->width_in_blocks; bx++)
{
blockptr_one = buffer_one[0][bx];
QVector<int> tmp;
for (int bi = 0; bi < 64; bi++)
{
tmp.append(blockptr_one[bi]);
}
dct_coeff.push_back(tmp);
}
}
}
// coantization table
j_decompress_ptr dec_cinfo = (j_decompress_ptr) &cinfo;
jpeg_component_info *ci_ptr = &dec_cinfo->comp_info[0];
JQUANT_TBL *tbl = ci_ptr->quant_table;
for(int ci =0 ; ci < 64; ci++){
quant_tbl.append(tbl->quantval[ci]);
}
return 1;
}
int main()
{
QVector<QVector<int> > v;
QVector<unsigned short> quant_tbl;
char *infilename = "your_image.jpg";
std::ofstream out;
out.open("out_dct.txt");
if( read_jpeg_file( infilename, v, quant_tbl ) > 0 ){
for(int j = 0; j < v.size(); j++ ){
for (int i = 0; i < v[0].size(); ++i){
out << v[j][i] << "\t";
}
out << "---------------" << std::endl;
}
out << "\n\n\n" << std::string(10,'-') << std::endl;
out << "\nQauntization Table:" << std::endl;
for(int i = 0; i < quant_tbl.size(); i++ ){
out << quant_tbl[i] << "\t";
}
}
else{
std::cout << "Can not read, Returned With Error";
return -1;
}
out.close();
return 0;
}
来源：https://stackoverflow.com/questions/4470107/extracting-dct-coefficients-from-encoded-images-and-video