For about six months I have been using Jasc Software's Paint Shop Pro version 6 to edit a series of old photographs. My one complaint about this excellent program was that it lacks sophisticated tools for correcting color balance in photographs. While the provided features certainly allow one to adjust the color balance, there are no tools in PSP6 to automatically analyze and balance the image. You are on your own in figuring out what to do.
On Friday, September 22 2000 I received an email notification of the release of version 7 of Paint Shop Pro. The advertisement noted that this version included a large number of new features, including several that were specifically related to photo correcting. I eagerly downloaded an evaluation copy and set to work trying things out. There are a large number of tools available in PSP7 and a few minutes of random experimentation convinced me that their effective use will require some serious study. I therefore created this document as a sort of lab notebook to record my findings.
So... this is an empirical study: I tried things, observed the results, and tried more things in pursuit of the best looking image I could get. I wrote this primarily because, for my own purposes, I needed a systematic way to record processing steps and compare results. Very little effort has been made to fully understand the precise nature of the transformations themselves, or the reasons why some things worked better than others. Even more to the point: I am not able to compare the photo correction tools of PSP 7 to other programs, particularly Adobe PhotoShop, because I don't own and have never used them. Nevertheless, I am making this document available to others in the modest hope that some may find it useful.
|The Sample Image||Automatic Color Balance|
|Automatic Fade Correction||More Work...|
My sample picture is a 300dpi scan of a 1968-vintage print that I took myself at Disneyland when I was 14 years old and had just graduated from 9th grade. The image is faded and somewhat discolored.
Click on this or the other images to see a much larger version
A selection of several images from that trip may be found at http://www.chipchapin.com/Dis68/index.html.
First, I used Histogram Adjustment. The histogram for Image 1 looks like this:
The PSP help file for Histogram Adjustment has many useful tips. I started by adjusting the High and Low values so that nearly 0.1% of pixels were at the high or low ends of the range. The image result with a low value of 13 and a high of 241 was as follows:
Better contrast, slightly brighter with slightly better saturation.
The next thing to do with Histogram Adjustment is to compress or expand the mid-tones. According to the help page, the bi-modal distribution of this image would suggest the use of mid-tone compression. The image below takes the one above and performs 10% mid-tone compression:
As one would expect, the 10% mid-tone compression nicely brings out image detail in the mid-range. But the smoother, less contrasty image seems more washed out.
I next tried mid-tone expansion. Here is the result of 10% mid-tone expansion:
Close inspection of the paddlers shows a much better image, in my opinion. However the image as a whole is too contrasty, and there is a loss of detail in the blacks.
The loss of black detail suggests that scaling up the low point may have been a mistake. I redid the whole sequence, leaving the low point untouched, and got the following:
High scaling and 10% mid-tone expansion. As expected, the detail in the blacks is better than the one above. However the image as a whole is somewhat more washed out.
Note: to avoid maintain consistency with the other examples above, the high-scaling and mid-tone expansion were done as separate sequential operations. However, in practice it would be possible to perform both at once.
Arguably, we may be better off without any mid-tone compression or expansion. I'll use Image 2 (high-low scaling, but no mid-tone correction) as the basis for the next batch of work.
Let's try Automatic Fade Correction, the simplest transformation, on the original image (Image 1). Automatic Fade Correction has a single "amount of correction" parameter. I used the default value of 45.
Automatic Fade Correction applied to original image.
The white margin has been corrected to nearly a true white, however the overall bluish cast is very unpleasant.
Next I'll try performing Automatic Fade Correction on Image 2 (the original with high-low scaling).
Automatic Fade Correction applied to Image 2.
The result is noticeable better, both in contrast and in color, than Image 6, however it still has the bluish tint.
The obvious conclusion is that Automatic Fade Correction by itself is no solution for this image. However it is encouraging to see what a nice job it did in correcting the whites. Later on, I try using the output of AFC as input for the Automatic Color Balance tool.
Next experiments are with the Automatic Color Balance feature, show below with its default values. As can be seen by the preview pane, the default values are going to yield an excessively bluish result (source is the original image).
The PSP help file suggests changing the color temperature to correspond to the original type of lighting, and using a higher value for very bright sunlight. However, a higher temp makes the image more blue. In our case, I found that a color temp setting of 4100K resulted in fairly natural colors. Here is the original image after processing with Automatic Color Balance, temp of 4100K, strength 30, and without "remove color cast":
Automatic Color Balance applied to original image at
Strength = 30.
These colors are the best yet, with realistic whites, some blue in the sky, and at least some greens.
Enabling "Remove Color Cast" loses on this picture. It doesn't do much, but does seem to reduce the greens. However fiddling with the strength value is interesting. Image 8a (above) is at strength=30. Shown below are the same image processed with strength=50, 65, 85.
|Image 8b||Image 8c|
Images 8b, c, d.
Automatic Color Balance applied to original image at
Strength = 50, 65, 85
I find the colors most natural with strength set to 30 or 50, though the higher strengths do create a more vibrant image.
Let's try ACB with Image 2 (after high-low scaling) as a base, using strengths 30 and 50.
|Image 9a||Image 9b|
|Automatic Color Balance applied to Histogram Adjusted image (Image 2) at strengths 30 and 50. Temperature = 4100K.|
For some reason these images are significantly warmer than their counterparts applied to the original image. The whites are less white, and the greens are much less natural. I decided to try increasing the color temperature and found better results at 4600K. The images below
|Image 10a||Image 10b|
|Image 10c||Image 10d|
Automatic Color Balance applied to Histogram Adjusted image (Image 2) at strengths 30, 50, 65, 85, 100.
Temperature = 4600K
I first ran strengths 30-85. I found that I liked 85 the best, with purer whites and less of yellowish cast (most noticeable in the greens and the water). So I also ran strength=100 (Image 10e), but to my eye the difference with 85 is not discernable.
The next experiment was to try using Automatic Color Balance to correct the bluish result of Automatic Fade Correction. Starting with Image 7, I kept the color temperature at 4600K and tried different strengths:
|Image 11a||Image 11b|
|Image 11c||Image 11d|
Automatic Color Balance applied after Automatic Fade Correction at strengths 30, 50, 65, 85, 100.
Temperature = 4600K.
I like these images very much. Surprisingly, I find it nearly impossible to visually discern any difference between Images 11a-e for the different strength values. However, differences do exist, as can be seen by a comparison of these two histograms:
Histogram of Image 11a.
Histogram of Image 11e.
The spikiness of 11e strikes me as unrealistic, even though I can't perceive it (at least not on my monitor). So I decided to use 11a as the basis for further work.
So which is better, the "Image 11 group" or the "Image 10 group"? I made a close comparison between Image 11a and Image 10d and observed the following:
So far, Image 11a is my favorite. To me it looks very natural and is in every way superior to the original. For easy comparison, here is the original Image 1 side by side with Image 11a:
|Image 1 - Original||Image 11a - Best so far|
To my eye, Image 11a looks fairly natural for very-late afternoon fading sunlight. Whether this is an accurate depiction of the original scene, I simply cannot recall, but it is quite possible.
I began to wonder whether the large amount of white and near-white appearing in the border might be skewing my results. I'd like to go back to the Histogram Adjustment and scale the "High" to a lower value (ie. clip more of the whites). An alternate approach would be to trim off the white margin before processing. In this case I left it on because I wanted to see what affect the processing would have on this nominally "white" region.
Also, I wonder what would be the affect on subsequent processing if I used mid-tone compression (i.e. Image 3)?
Both of these should be pursued independently.