topic Re: RGB2Gray round off? in Intel® Integrated Performance Primitives

RGB2Gray round off?

yaddayadda22 — Tue, 30 Jan 2007 08:48:26 GMT

Hi,

According to the image processing documentation, functional results are often "rounded to the nearest integer". I am finding this not to be the case. For example, the following code demonstrates what I'm finding.

BYTE* boo = new BYTE[ 1 * 3 ];
IppiSize boo_roi;
boo_roi.height = 1;
boo_roi.width = 3;
boo[0] = 113;
boo[1] = 134;
boo[2] = 115;

BYTE* bood = new BYTE[ 1 * 1];
ippiRGBToGray_8u_C3C1R( boo, 3, bood, 1, boo_roi);

printf("TEST RGB TO GRAY on (113,134,115) = %d yet math = %.4f ",bood[0],
0.299*113 + 0.587*134 + 0.114*115);

Output is the following :
TEST RGB TO GRAY on (113,134,115) = 125 yet math = 125.5550

So if the floating point value of one-pixel rgb image if 125.550, shouldn't the nearest integer be 126, not 125?

Thanks
Matt

Re: RGB2Gray round off?

Vladimir_Dudnik — Mon, 05 Feb 2007 06:56:05 GMT

Hi Matt,

you are right, it was an issue in IPP 5.1. We have fixed it in IPP 5.2 beta

Regards,
Vladimir

Re: RGB2Gray round off?

Vladimir_Dudnik — Tue, 06 Feb 2007 04:40:11 GMT

Ops, just double-check with developer, the more correct statement is that we provide +/- 1 precision in that function because of performance tradeoff. Please let us know if this is critical issue for you.

Vladimir