Chromatic aberration (CA) is an unwanted photography effect caused by the lens which focuses different light frequency at different distance. The effect are red, blue, yellow or cyan fringing at the corner of an image.
RawCA calculates lens' CA from a camera raw file. It works by searching for the largest correlation coefficient of different channels while scaling (stretch/shrink) the channels. RawCA uses raw files instead of jpeg files because CA can be reduced/eliminated by the demosaicing process which is part of the raw-to-jpeg conversion. Another purpose of this project is to build a database of different lens' CA.
Generally speaking, CA is a property of a lens, thus CA only depends on lens. However, since the distortion of different color channel is not a linear scaling (It can be a complicated function), and the way I measure CA is a linear ratio, sensor size also affect CA. Consider an image which has little CA at the center area but strong CA at the corner, cropping the image (reduce sensor size) reduces the CA. Note that only sensor size (millimeter) affects CA, not resolution (mega pixel). As long as the sensor size is the same, resolution doesn't affect CA, because my measure of CA is a ratio (E.g. the red channel is 0.01% smaller than the blue channel.) instead of number of pixels (E.g. the red is 2 pixels smaller than the blue at the corner.). I know some lens benchmarks use number of pixels, but I don't. For exmaple, the same lens can have different CA on a full frame camera and an APS-C camera, but the same CA on 8MP and 12MP APS-C cameras.
| factor | affects CA? | affects accuracy of the calculation? |
|---|---|---|
| shutter speed | no | No as long as 1. the camera is stable; and 2. exposure is correct. |
| aperture | yes for some lens, but not serious. TODO: larger or smaller? | yes. I recommend using medium aperture, because too large aperture makes image less sharp, too small aperture causes diffraction which also makes image less sharp. When image is not sharp, calculation of CA is less accurate. |
| lens | yes of course | N.A. |
| focal length in zoom lens | yes | N.A. |
| filter | very little effect unless it's a very crappy filter | Theoretically filters introduce flare, so it's less accurate. But usually it doesn't matter. |
| camera | No as long as the size (not resolution) of the sensors are the same. See the discussion on sensor size and resolution above. | yes. The larger the resolution, the more accurate. |
| ISO | no | yes. The less the noise (i.e. smaller ISO), the more the accuracy. |
| flash | no | no |
| subject/scene | no | yes. See discussion below. |
| raw/jpeg | no | yes. (1) The demosaicing process reduces or eliminates CA. (2) JPEG contains very little color information due to down sampling Cb and Cr, but the color is what we want. (3) Usually cameras slightly crop the image when saving into JPEG, but the corner is what we want. |
| 12 or 14 bit raw | no | Theoretically the larger the dynamic range, the more accurate. But in practise, I believe the 2 additional bits are mostly noise. |
| cropping | If image center is preserved, it's like reducing sensor size (see camera above). If image center is not preserved, it's hard to say anything on the cropped CA. | N.A. |