Cameras like the NEX-7, with their high resolution sensors and low flange distances, return to photography the kind of flexibility in lens selection associated with view cameras. That flexibility is not without cost. There is a loss in automation, which is no big deal to most of us. The biggest effect is that the responsibility for control over the ultimate image quality now rests squarely with the photographer, since the camera manufacturer doesn’t know whose lenses you are going to use, and the lens manufacturers don’t know what camera you’re going to put their product on.
The “purple corner” effect is an example of this. Sony is not going to help us out. Leica is not going to help us out. We have to figure out what to do to get the most out of these hybrid imaging systems that we are constructing on an ad hoc basis.
The solution to purple corners can be in quick and dirty Photoshop techniques like the ones I described in the last post, or it can come in the form of programs written to accomplish the task, like Cornerfix (thanks to Rich for giving me the pointer to that program in a comment to the previous post). Both solutions, and any other that I can think of, require making test images with the camera/lens combination for which corrections are desired.
Therefore, I’m writing this post on making test images to correct color and luminosity errors. It should be applicable to just about any camera and lens combination, but all my experimentation has been done with the NEX-7, and Sony, Leica, and Zeiss lenses.
I’ve already done made some test images, and reported the results in the previous post. However, I was unsatisfied; I thought I had to find a more accurate and more repeatable way to get a capture of a constant neutral Zone V field. I experimented with illuminating a wall, measuring the result with a spot meter, and was unhappy with the level of smoothness of illumination I could achieve.
I tried a different approach: I took two pieces of paper, and folded them over so I had four thicknesses. I put the paper over the front of the lens, aimed the camera directly at the sun, and tripped the shutter. I compared the results with pictures of my not-quite-evenly-illuminated wall. Using the paper diffuser produced much more even and repeatable results, but seemed to understate the amount of magenta in the corners. As I played with various lenses, I noticed something else. Wide angle lenses exhibited mottled illumination of the image plane, with the Tri-Elmar behaving particularly badly in this regard, possibly because of its complex optical design. I’ve never seen this effect in actual photographs. Can it be caused by flare or some other anomaly associated with my diffuser trick? Yes it was. The wall pictures don’t show the mottling.
As an aside, there is a theoretical problem with flare when using lenses designed for full frame sensors on cameras with APS-C sensors like the NEX-7. The full frame lenses will spray light far outside the small sensors, and some of that extra light may bounce around the inside of the camera and end up on the sensor. I don’t know how much of a problem this is with the NEX-7, but I’ll try to think of a way to test it.
Back to the test image. Now I was suspicious of the whole diffuser shortcut. I decided to go back and find a way to make my photographs of the wall more accurate and more repeatable. The classic way of making copy slides of photographic prints is to arrange them on a wall, with two lights at 45° angles on the right and left. That arrangement redirects any specular reflections back at the other light and away from the camera lens. It also allows precise control of the color temperature of the light source. It isn’t clear that such precise control is necessary, since the image is color balanced after the exposure, but it prevents possible capture metamerism.
The sources of error were the evenness of illumination of the wall by the light, and possible reflectivity differences in the wall itself. I figured that I could minimize the effect of both of those by averaging several exposures. Ultimately, it turned out that four exposures with the camera at various angles – horizontal, vertical to the right, vertical to the left, and upside down horizontal – were sufficient for excellent repeatability.
There’s another source of error that I had to think about: subject distance. Would one correction work for a whole range of subject distances for the class of lenses that needed the purple corner correction? Would it be better to focus the lens on the test wall, or focus it at infinity? Focusing it on the wall would move the nodal point of the lens further away from the film, reducing the purple corner effect, likely by a negligibly small amount.
I conducted some experiments with the 24 mm Elmar. Focusing on a wall 2.5 feet away and making an identical exposure with the wall at the same distance and the lens focused on infinity produced images with some L* differences, but with the substantially the same a* and b* values. I couldn’t do the obvious next test, with the lens focused on infinity and the wall far enough way that the wall was within the sharp region by virtue of the depth of field because I didn’t have a big enough, evenly lit wall.
I compared the averaged test images of the wall with the photos I had made of the overcast sky earlier. Both showed similar corner color shifts, but the averaged wall pictures were smoother, and didn’t have that light region at the top. This made me think I was on the right track.
A word about averaging. I did it by putting all four exposures into separate layers in Photoshop, using a linear version (gamma equals one) of the ProPhoto RGB color space, using normal blending mode, setting the opacity of the second layer from the bottom to 50%, the opacity of the third layer from the bottom to 33%, and the opacity of the fourth layer from the bottom to 25%. It looks like you can achieve the same results in regular ProPhoto RGB by going into the color settings, checking the box that says “Blend RGB color using gamma:” and accepting the default unity value for gamma.
Compositing the pictures for averaging was made significantly more challenging because I can’t figure out how to turn off the automatic image rotation “feature” in the NEX=7. Can anybody help me with that? At least it doesn’t work when the camera is upside down.
I made a set of test exposures at f/8 with the f/3.8 Leica Elmar. Here’s the composite:
Setting the eyedropper to do a 51×51 pixel average, I measured the center at L*=49, a*=0, b*=0. The upper right corner was L*=35, a*=5, and b*=-8. The bottom right corner was L*=34, a*=4, and b*=-7. The lower left corner was L*=36, a*=2, and b*=-5. The upper left corner was L*=34, a*=2, and b*=-3.
The next step is producing test images for more lenses. This will not only tell me which ones need correcting, but will be the first step in developing a correction process.