5

What is it?

"Large Areas Spatial Crosstalk" is the terminology used by the IEC standard for scanners, 61966-8. It refers to the effect of nearby colors shifting the scanned colors towards the neighboring colors.

What problems does it cause?

The effect can produce large differences in the colors of scanned photos which makes it sometimes impossible to scan a photo and get a colorimetrically accurate image file. These aren't often noticed in complex photos but some are quite susceptible. A scanned picture of a hot air balloon against a bright sky will tend to lighten the brighter sky more than the balloon.

Accurate reproduction is highly desired for, inter alia, precision duplication and retaining historical images.

Here's a scanned image from a moderately high end, V850 Epson Scanner. It's large, shifted L* values are labeled underneath each circle. The three, circular, patches on the left were printed the same as the three on the right. The lighter ones were RGB(240,240,240) while the darker ones were RGB(118,118,118). The circles surrounded by white have scanned L*s about 6 higher than the same ones on the right. This shift is from "Large Area Spatial Crosstalk."

Scan with Large Area Spatial Crosstalk

This problem is caused by light reflecting off nearby parts of a document, getting scattered around and bouncing off the translucent, illuminating white rails on either side of a horizontal aperture through which the CCD scan is captured. It's as if the Lux levels on the right side white surround area were boosted over 20%, the increase in light needed to increase L* from 85.7 to 92.7.

Question: Are their any reflection desktop scanners that have addressed this issue in their design and/or is there any software that corrects or is looking to correct this defect?

  • Though it's probably not practical, if one were to disassemble and paint the inside of a scanner black, would large-area spatial cross-talk be expected to be reduced? What about multi-pass scanning at reduced exposure/light levels? – xiota Nov 23 '18 at 8:17
  • At first blush I thought the same thing. And it may be a help when scanning transmission media (film) but glare is minimal there. Reflective media's problems is the light that reflects off the frosted, white surfaces that are used to diffuse the LED light. Since the frosted surfaces are both pretty reflective and are very close ( a 1-2 cm's) virtually all the light that makes it's way back to the spot being scanned is light reflected off of those and they have to be translucent, and hence fairly reflective. Because they are so close, the reflected light can be quite a high percentage (>20%). – doug Nov 23 '18 at 18:26
  • In photography, this is referred to as the "adjacency effect" illustrated by a woman appearing with a decidedly ruddy complexion—due to the large percentage of the image being a red brick wall behind her. – Stan Nov 28 '18 at 20:14
  • @Stan - Haven't seen the term applied to reflection scanners but transmission film scanners see it. Some is due to scanner light bouncing and becoming colored within the film itself and some is due to the chemical processing and is part of the film being scanner. Both are very short range effects compared to the multiple cm spatial crosstalk problem of reflection scanners. Seems the term is often used re satellite images where it's caused by the atmosphere. That's actually more similar to the reflection scanner problem. – doug Nov 28 '18 at 22:08
3

This is a partial solution in that it only works for the Epson V850.

In the case of the V850, I've put together a program in C++ on github tailored to the Epson V850 which models the neighboring, reflected light and subtracts it from the image. However, this uses a model specific to the Epson V850. It corrects most of the large area spatial crosstalk as you can see from the attached image's improved L* value consistency. Unfortunately, this program uses a model that is specific to the V850 design and does not work well on other scanners with differing aperture designs. It does, however, work quite well on a variety of media such as matte, glossy and semigloss. There is some variation scanning different paper surface types but it is surprisingly small. At least with the V850. I'm looking for a more general solution.

Crosstalk Corrected Image

  • 1
    If you're interested in making your code even better, may I humbly suggest asking for critique over on Code Review? It looks to be compatible with the CC-BY-SA required to reproduce it on Stack Exchange, and you could get useful advice (the great thing about advice being that you can choose whether or not to act on it, of course!) It might be possible to further parameterize the code to work better with different scanner geometry, for example. – Toby Speight Nov 23 '18 at 8:36
  • @TobySpeight Interesting place. My code is constructed somewhat unusually, with the RGB plane broken into separate planes. This provides better cache locality because the math involved is mostly specific to each plane. But it makes the structure look quite weird. In any case the area in need of improvement is constructing the 2D reflectance set. It's a set that models how much light is reflected from neighbors at offsets of x,y. My main tool for that is Matlab using a randomly placed set of 3mm black/white squares. Might be something for the engineering or physics groups. – doug Nov 23 '18 at 18:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.