weird image while trying to compress YUV image to jpeg using libjpeg - c++

Question

I am using Qt,OpenCV and libJpeg to compress a YUV422 image, but the output is not correct.
if I convert the yuv to rgb and then compress, I get a correct output, but what I understand that Jpeg is using YUV internally , that's why I want to remove redundancy.
Here is my code:
bool ipl2jpeg(IplImage *frame, unsigned char **outbuffer, long unsigned int *outlen) {
unsigned char *outdata = (uchar *) frame->imageData;
struct jpeg_compress_struct cinfo ;
struct jpeg_error_mgr jerr;
JSAMPROW row_ptr[1];
int row_stride;
*outbuffer = NULL;
*outlen = 0;
cinfo.err = jpeg_std_error(&jerr);
jpeg_create_compress(&cinfo);
jpeg_mem_dest(&cinfo, outbuffer, outlen);
cinfo.image_width = frame->width;
cinfo.image_height = frame->height;
cinfo.input_components = frame->nChannels;
cinfo.in_color_space = JCS_YCbCr;
jpeg_set_defaults(&cinfo);
jpeg_set_quality (&cinfo,100 , true);
jpeg_start_compress(&cinfo, TRUE);
row_stride = frame->width *2;// frame->nChannels;
while (cinfo.next_scanline < cinfo.image_height) {
/* jpeg_write_scanlines expects an array of pointers to scanlines.
* Here the array is only one element long, but you could pass
* more than one scanline at a time if that's more convenient.
*/
row_ptr[0] = &outdata[cinfo.next_scanline * row_stride];
(void) jpeg_write_scanlines(&cinfo, row_ptr, 1);
}
jpeg_finish_compress(&cinfo);
jpeg_destroy_compress(&cinfo);
return true;
}
What is wrong ??
Here is the YUV image

Answer 1

the image file provided (frame-3.raw) is 640x480 in YUYV format
the following code
JSAMPROW row_pointer[1];
row_pointer[0] = row_buf;
while (cinfo.next_scanline < cinfo.image_height) {
unsigned i, j;
unsigned offset = cinfo.next_scanline * cinfo.image_width * 2;
for (i = 0, j = 0; i < cinfo.image_width*2; i += 4, j += 6) {
row_buf[j + 0] = buf[offset + i + 0]; // Y
row_buf[j + 1] = buf[offset + i + 1]; // U
row_buf[j + 2] = buf[offset + i + 3]; // V
row_buf[j + 3] = buf[offset + i + 2]; // Y
row_buf[j + 4] = buf[offset + i + 1]; // U
row_buf[j + 5] = buf[offset + i + 3]; // V
}
jpeg_write_scanlines(&cinfo, row_pointer, 1);
}
to be put in between jpeg_start_compress() and jpeg_finish_compress() creates a JPEG file correctly without color space conversion (but with upsampling)
jason_s's code looks correct as well

Answer 2

I don't think libjpeg will accept subsampled image data as input. So, you'll have to un-subsample it first.
You could allocate a temporary row buffer, and convert each row, something like this:
for (i=0; i<frame->width; i++) {
tmpbuf[i*3+0] = outdata[cinfo.next_scanline * row_stride + i*2];
tmpbuf[i*3+1] = outdata[cinfo.next_scanline * row_stride + (i-i%2)*2+1];
tmpbuf[i*3+2] = outdata[cinfo.next_scanline * row_stride + (i-i%2)*2+3];
}
row_ptr[0] = tmpbuf;