topic JPEG 2000 compression in Intel® Integrated Performance Primitives

JPEG 2000 compression

jboaz123 — Sun, 07 Feb 2010 11:42:49 GMT

Hi,

I am trying to compress the attached file using picnic application with the following parameters:

Files of type:JP2 Format (*.jp2)

Lossless

WT53

Use MCT checked

when I load save result I get gray image.

Problem not occurwith lossy.

Please advice

10x

JPEG 2000 compression

Thomas_Jensen1 — Sun, 07 Feb 2010 15:33:51 GMT

I took a look at the dicom file. It is not properly structured:

Missing tag: SAMPLES_PER_PX (0028:0002)

It is bad to not indicate how many channels a pixel has.

Missing tag: PHOTOMETRIC (0028:0004)

It is bad to not indicate how the pixels are structured.

Also missing tag: PIXEL_SPACING (0028:0030), should indicate size of a pixel.

Further, Bits Stored = 14

The IPP Dicom decoder does not use two very important tags: BITS_STORED (0028:0101) and HIGH_BIT (0028:0102). The decoder assumes the image is 16 bit, and thus with 14 bits, the whitest pixel has value 2^14-1, which in 16 bit space is 1/4 of white, or, dark gray.

If you want to fix this problem, you must modify the IPP Dicom decoder to return tags Bits Stored and High Bit.

Using that, you can add code to Left_Shift all bits 16-HighBit (16-14 = 2 -> 14 bit becomes 16 bit).

Actually, although your Dicom file is 14 bit, only 11 bits are used, thus even when upscaling the specified 14 bit to the full 16 bit, that result will display as a very dark gray image. To fix that, you need to disregard the specified 14 bit and "detect" the actual number of bits (11), and then do the upscaling.

Intel could consider adding code to the Dicom decoder sample, to upscale the image from <16 bit (9..15 bit) to full 16 bit, when returning a 16 bit image. Motivation: The Dicom file specifies 14 bit, so 2^14-1 = white. Therefore in the context of the Dicom decoder, the 14 bits *must* be considered full range, and then, since it returns a 16 bit image, it must remap 14 bit white to 16 bit white.

JPEG 2000 compression

Vladimir_Dudnik — Sun, 07 Feb 2010 17:13:10 GMT

Hi Thomas,

thanks for loking into problem, your comments looks reasonable. I think we will adopt DICOM parser according your advices.

Regards,
Vladimir

JPEG 2000 compression

jboaz123 — Mon, 08 Feb 2010 08:15:34 GMT

Hi,

I attached correct 8 bit DICOM file and the jpeg 2000 lossless compression result image Compredded.jp2.

You can notice the wrong result.

JPEG 2000 compression

Sergey_Ryadno — Mon, 08 Feb 2010 12:00:17 GMT

Hi, jboaz123

I found reason of this problem. In case of Lossless mode for J2K image saving variableparam.quality always equal to 0. This bring to encoding image to zero size. To fix this issue you need to change one row in filesavedlg.cpp. In functionvoid FileSaveDialog::updateJpeg2000Buttons(void) change:

m_jpeg2000RatioSlider.setValue(0);tom_jpeg2000RatioSlider.setValue(100);

JPEG 2000 compression

jboaz123 — Mon, 08 Feb 2010 14:09:48 GMT

Hi, sryadno

Thank you for your advice, I will check it.

JPEG 2000 compression

Thomas_Jensen1 — Mon, 08 Feb 2010 17:09:41 GMT

I can confirm that saving Jpeg2000 Lossless 12 bit grayscale always gives a 1KB useless file, in UIC samples version 6.1.3.052.

JPEG 2000 compression

Sergey_Ryadno — Tue, 09 Feb 2010 08:58:45 GMT

HiThomas,

the fix i described one post early remove this problem. i just test picnic with this changes on 12/16 bit grayscaleimage - and it works fine.