Color Bleeding

Dear Group

I have come across an issue with the color bleeding of image maps that I use with my Radiance renderings. I have the dual objective of achieving a physically accurate and asthetically pleasing rendering. It is my understanding that to obtain physically accurate attributes for a material associated with an image map, one should normpat'd the image and then set the appropriate rgb reflection for its associated material. The problem I have come into is that the color bleedng seems excessive. Here are some examples:

For the wood floor image map in this rendering, the image map was not normpat'd and a rgb reflection for the floor (plastic) was set at .4 .4 .4. The floor image map brightness is a bit darker than what I would expect:

For this rendering, the wood floor image image was normpat'd. The red green blue option in colorpict was set to red green blue and a rgb reflection for the floor (plastic) was set at .4 .4 .4. Notice the level of color bleeding in the rendering. Also, the brightness and saturation of the image map is much higher than the pre-normpat'd image map:

In this rendering, the wood floor image map was not normpat'd and a rgb reflection for the floor (plastic) was set at 1 1 1. I know not how physically accurate the floor material is reacting but the brightness of the floor material is what I think I would expect:

In this rendering, the wood floor image map was normpat'd, the red green blue option in colorpict was set to clip clip clip and a rgb reflection for the floor (plastic) was set at .4 .4 .4:

In this rendering, the wood floor image map was normpat'd, the red green blue option in colorpict was set to grey grey grey and a rgb reflection for the floor (plastic) was set at .4 .4 .4:

In this rendering, the wood floor image map was normpat'd, the red green blue option in colorpict was set to noop noop noop (I think this was recommended in RWR) and a rgb reflection for the floor (plastic) was set at .4 .4 .4. This rendering and the floor is very saturated.

I guess this is the rundown. Any suggestions?

Thanks

Marcus

Hi Marcus,

In brief, I think that 40% reflectance for a wood floor is too high. An off-white wall might have a 40% reflectance, but most wood floors would be closer to 20%. If you use macbethcal for your floor capture, then it gives you back true reflectance values. In such cases, I would not pass the pattern trhough normpat.

Further comments below:

From: "Marcus Jacobs" <[email protected]>
Date: March 11, 2005 11:34:15 AM PST

For the wood floor image map in this rendering, the image map was not normpat'd and a rgb reflection for the floor (plastic) was set at .4 .4 .4. The floor image map brightness is a bit darker than what I would expect:

http://pg.photos.yahoo.com/ph/marcdevon/detail?.dir=/b3e9&.dnm=d98b.jpg&.src=ph

Here you are multiplying your captured reflectance from the image by 40%, which makes it 40% of what it should be (most likely). Too dark, as you say.

For this rendering, the wood floor image image was normpat'd. The red green blue option in colorpict was set to red green blue and a rgb reflection for the floor (plastic) was set at .4 .4 .4. Notice the level of color bleeding in the rendering. Also, the brightness and saturation of the image map is much higher than the pre-normpat'd image map:

http://pg.photos.yahoo.com/ph/marcdevon/detail?.dir=/b3e9&.dnm=e6b4.jpg&.src=ph

Probably because 40% is too high a reflectance for your floor.

In this rendering, the wood floor image map was not normpat'd and a rgb reflection for the floor (plastic) was set at 1 1 1. I know not how physically accurate the floor material is reacting but the brightness of the floor material is what I think I would expect:

http://pg.photos.yahoo.com/ph/marcdevon/detail?.dir=/b3e9&.dnm=adb6.jpg&.src=ph

If you were using macbethcal to create your floor pattern, this is the technique I would recommend (using the functions red() green() and blue() as the first three string arguments in your colorpict).

In this rendering, the wood floor image map was normpat'd, the red green blue option in colorpict was set to clip clip clip and a rgb reflection for the floor (plastic) was set at .4 .4 .4:

http://pg.photos.yahoo.com/ph/marcdevon/detail?.dir=/b3e9&.dnm=3242.jpg&.src=ph

I assume you mean clip_r(), clip_g() and clip_b(). The function clip() by itself would always return the red channel, and you would end up with a gray-looking floor.

In this rendering, the wood floor image map was normpat'd, the red green blue option in colorpict was set to grey grey grey and a rgb reflection for the floor (plastic) was set at .4 .4 .4:

http://pg.photos.yahoo.com/ph/marcdevon/detail?.dir=/b3e9&.dnm=36c5.jpg&.src=ph

In this rendering, the wood floor image map was normpat'd, the red green blue option in colorpict was set to noop noop noop (I think this was recommended in RWR) and a rgb reflection for the floor (plastic) was set at .4 .4 .4. This rendering and the floor is very saturated.

http://pg.photos.yahoo.com/ph/marcdevon/detail?.dir=/b3e9&.dnm=d1fc.jpg&.src=ph

Again, I don't see how using noop() would give you this image, as noop() expects a single argument, and if given three (r,g,b), will return the first.

I guess this is the rundown. Any suggestions?

Again, I suggest using macbethcal to capture your image, and don't run normpat unless you have a measured reflectance value to use.

-Greg

Hi Marcus,

one hint concerning the expected appearance:

I assume that the image map you use for the floor originally is a photo (TIFF, JPG, TGA or whatever), converted to a Radiance pic. In the conversion process a gamma correction is applied (unless you set the option -g 1.0).

The gamma correction makes the images appear darker, means if you click on your photo in an image editing program, and it gives you, say, 0.6/0.5/0.1 as rgb (or 153/127/25 in the 0-255 range), the corresponding r/g/b coefficients will be lower after the import into Radiance.

Only if you set -g 1.0 (e.g. in ra_t16 -r -g 1.0 test.tga test.pic) and then map the image onto a plastic with rgb = 1/1/1, you'll get the original r/g/b values form the picture (regardless if they're physically meaningful or not), means the appearance will be the same as if you use a 'plastic' with rgb = 0.6/0.5/0.1 directly.

-cb

PS
nice images…is this the room in the registry office where Prince Charles and Camilla are going to hold their wedding??? Especially the pink version would suit perfectly for it !!