Canon Digital Rebel Color Resolution Tests

Page last edited Nov 19, 2005, 1:45 pm PST

The following images were shot to evaluate the effective resolution of a Canon Digital Rebel camera (3072 x 2048 pixels) when dealing with single-color light.

Summary:

Lens: Sigma 18-125mm at 18 mm, f/11
Illumination: fluorescent
Color filters: Kodak Wratten gelatins, in front of the lens, Numbers 29 (red), 61 (green), 47B (blue).
File format and processing: shot in Canon raw (.crw), converted to RGB using Photoshop CS raw converter with all settings default except Sharpness 0 and Color Noise Reduction 0.
For display: colored images (red, green, blue) are converted to gray scale by replicating the appropriate channel.
Update (Nov. 13, 2005): histograms of the converted RGB channels were added.
Update (Nov. 18, 2005): raw sensor data was extracted with dcraw -d. This data clearly shows the Bayer pattern. It also indicates that both the red and blue sensor positions are significantly sensitive to light getting through the green (#61) filter. This contradicts what was suggested by the histograms of the converted RGB channels, and helps to explain why there is only a slight resolution loss for green light, but significant for red and blue.

Major result: There is a slight resolution loss for green light as opposed to white, but a significant loss for red and blue.

However, the nature of the resolution loss is more like blurring than the block effect described at http://groups.yahoo.com/group/Microscope/message/27425 and illustrated in the links at http://groups.yahoo.com/group/Microscope/message/27430 . Presumably this is due to differences in the raw-to-RGB conversion algorithm.

Just for context, here is the full-frame view with no filters. We're going to look at the small block in the center of the resolution chart.

All of the following images have been blown up to 400% with nearest-neighbor interpolation in Photoshop.
The inset at lower right is the same image at actual pixels (100%).

Here is the central portion of the resolution chart. The order is 1) no filter, 2) #29 red, 3) #61 green, and 4) #47B blue.

Histograms:

To illustrate the effect of Sharpness and Color Noise Reduction settings, here are two no-filter images. On the left is with default settings (Sharpness=25, Color Noise Reduction = 25); on the right is with Sharpness=0, Color Noise Reduction = 0, same as shown above.

No filter, default sharpening and smoothing

Tiff downloads:
RedImage.tif
GreenImage.tif
BlueImage.tif
NoFilter Options0 Image.tif
NoFilter OptionsDefault Image.tif

Update Nov 18, 2005:

The raw-format CRW files were processed through dcraw (http://www.cybercom.net/~dcoffin/dcraw/) using the -d parameter to extract raw sensor data (no de-Bayering, no interpolation). The resulting files clearly show what appears to be the Bayer filter pattern. Here is the dcraw -d output for the red, green, and blue filters, applied as a Photoshop layer mask against uniform red, green, and blue tones to help visualize them.

Notice that the green filter produces significant response at all sensor positions.

Numerically, the responses in the upper-left Bayer quad are:

	Bayer Filter Position
Filter in front of lens		Red	Green	Blue
	Red	232	27	0
	Green	120	185	75
	Blue	10	68	181

These values are after conversion by IrfanView from dcraw's -d ppm output to 8-bits-per-color tiff. They may or may not represent actual sensor values, but should give an idea of relative responsiveness.