How does TCA look like?

Summary

One should not to confuse the color fringing that is caused by TCA with the true TCA itself. True TCA (color error) is extended (--> fringe width) and diluted (--> color de-saturation) by edge blurring. In other words: in "color fringe" color is caused by TCA and fringe is caused by lens optical softness.

While I made personally the confusion until recently, I would now, at least "literally", disagree with some of the most respectable authors such as Ken Rockwell for instance who wrote:

<< It (TCA) doesn't vary with aperture, although a lack of sharpness will hide it.>> In fact and on the contrary, the lack of sharpness makes it more visible especially at lower magnification viewing level.

An experiment shall demonstrate this provocative statement.

Introduction:

Transverse Chromatic Aberration (a.k.a. TCA) has been described by many knowledgeable authors in the wikipedias, for example in (1), (2), or in well documented articles such as (3) or (4), just to list few of them.

--> TCA (Lateral Color) is in the focal plane, caused by magnification difference function of light wavelength. Full stop.

We may then have learned why there is TCA. Then does that defines (or do we know) how TCA looks like?

As we have learned to recognize TCA because we generally feel necessary to correct it on the source images before stitching, I would therefore answer positively.

Most of our favorite lenses used for panographic purpose and especially the fisheye kind, suffer more or less from this inevitable plague. The larger is the Angle of View, the stronger is TCA along the radius of the circle of the image: this is what we all have experienced generally.

With some more experience, we even know to not confuse TCA with other similar type of color aberration. The famous "Purple Fringe" is often classified also in the Chromatic or Color type of Aberration. I know of some CA correctives that used to correct only this type of CA (BTW that is unrelated with lens optics) and calling the TCA "difficult to remove" before finally introducing a tool designed to attempt in correcting also the TCA in a recent beta version.

Another defect, Longitudinal Chromatic Aberration (LCA) is more difficult to assess, but it is often the reason why there is some slight aberration visible near the center of the image. For higher end digital cameras, it has generally much more moderate effect than TCA. This is fortunate because there is little chance to be able to correct LCA. BTW: It generates a part of the above mentioned softness of the lens and reduces contrast.

So with all those material about TCA that we can read, with so many different tools that are supposed to fully correct TCA, we know how to recognize TCA before getting rid of this optical error, don't we?

In fact I recently discovered that I did not know personally what was the true TCA appearance on my image! This is why I shall try to share this with the reader.


Experimenting to view what does pure TCA looks like:

Let's recap what the wiki.panotools.org tells us on "how to distinguish":

Then some lines further down:

... the best subject for CA determination is composed of neutral greys, black and not overexposed white.

If this is the key to experimentally determine TCA with assurance, then let's make such an artificial image for that sole purpose:

Click on the image to view it full scale
You may download the full scale image to continue with the experiment by yourself.

Three ladder patterns span from the center to the lower left corner of the image:

  • A = equally spaced B &W steps with sharp edges
  • B = equally spaced B &W steps with edges that are blurred increasingly toward the corner (radial lens blur)
  • C = equally spaced B &W steps with edges that are blurred decreasingly toward the corner

Comments about this image:

Below is an enlarged view of the lower left corner of the source image:

Click on the image to view it full scale

For the following experiment steps we are going to apply a correction of the Transverse Chromatic Aberration. As there is no such Aberration on the source image, the "correction" shall produce some!

I have personally used the Adobe Camera Raw 4.1 for that purpose. This newest version supports TIF and JPEG images as well as RAW format. The reader-experimenter may apply radial shift correction with another correction tool if he wishes to do so.

So, we shall simply move the Red/Cyan TCA correction slider to the right to +40 and the "corrected image" looks now like this:

Click on the image to view it full scale

Below is an enlarged view of the lower left corner of the source image:

Click on the image to view it full scale

Observations:

From its definition and physical nature, the TCA (artificial or not) has to be the same at radial equidistance from the center on the image. It is therefore the same for the three ladder pattern. The width of the pure TCA fringe is therefore the width if the fringe that borders the sharp edges of A on top of the two others. While I have read this wrong (qualitative) statement quite often, TCA width is not the width of the soft part of the blurred steps B. The color of the colored fringe is much more saturated along the sharp edge than in the blurred edge: the colored is "spread" over a much larger area: this is the typical look on real images (see at the bottom of the page).

The actual width of TCA is much smaller than the APPARENT fringe that is in fact about the "width of the softness" of the lens in the comer. There is about a 1/10 ratio. This also implies that full correction of the TCA of a "soft" lens shall not correct the optical softness (while it helps in reducing some blur in the details of the image, though): as it is about equivalent as the reverse operation from this "corrected" image back to the source image by application of +40 of Red/Cyan correction level in ACR 4.1.


Real life example:

We may estimate the width of the magenta colored fringe at the lower left corner to be about 13-15 pixels in width (i.e. 13 x 36 / 4386 = 0.106 mm on my EOS 5D sensor)

Measured true TCA (Graph):

You may see this graph in full scale PDF format in two different versions that differ by the Y-axis unit:

or

Conclusion:

On a typical fisheye lens, the apparent color fringe width is about ten times larger than the real TCA extent (TCA being the shift of the red and blue channels with respect to the green channel).

The confusion that has commonly taken place in the literature is however fortunately of no consequence. The correction is usually done with tools of which the settings are with arbitrary units that have only relative values. The unit is generally neither pixel, nor mm, nor % of distance: it simply is absolute unit of correction for TCA for the blue or the red channel for this specific tool.

What really counts is the ability to remove the Lateral Color during post processing of the real camera source image. This is another story (stay tune).

I hope to have clarified this colorful but blurred subject:-)


Michel Thoby

30 July 2007