The method of blurred differences as applied to the task of simultaneous correction of several colors.

Imagine that you have scanned the negative with a high resolution. And you have made it by setting the mark “Negative” at the interface of the scanner. Surely you have a familiar picture with ragged colors.

The negative is designed for the usual photo printing where the thin effects of overlay colors make just another presentation of the color when examining the print in the reflected light rather than on slides. Scanner software tries to improve the brightness, tone and intensity of colors. But when working with colored grain pattern of the scanned image, we face the unsolvable problem. As greenery along with high resolution consists not only of green grains but yellow, blue and brown also. And each of them is quite far from the average green which scanner’s driver tries to improve grasping just a small piece of the image, part of grains. Others not counted. And in order to achieve the required changing of the color in on average, the program takes the green color to the endless expanse of the colored space enormously. And this turns to be a disaster: only rags are left from our greenery…

But this idea hasn’t hit you up yet. You remove the mark and go one scanning like a slide. I hope you didn’t forget to make a mark for 16 bits for the channel? Otherwise even if you had set up the prior settings according the balance, you still can expect the significant color correction. And the absence of the enough amount of the luminance range brings to the irretrievable adhesion (pasteurization).

After the scanning, already in Photoshop, you have set up the ranges in channels and scales at Levels and Curves. And you have discovered that even along with the right balance on grey, you get something wrong: the greenery is still too bright  of some light-green color, the roads are of twee cream color and the blue sky, very beautiful on the photo, is somewhat colorless.

And still the same work is waiting for you, and some “Super-Cool-Scan” has already produced the latest bits weakly. And you go one with Hue/Saturation, Replace color…

What you get is not much beautiful. So you recall the phrase you have read some day: “Smooth over the colored component of the picture”. And really, the already dying picture starts soften and something is trembling there. It’s quite much easier to manage the colors. But! You discover that even before the beginning of the correction the entire image is already desperately spoiled: the small colored spots have assumed the background color, there is no any change of colors, and someone’s lips are of yellow tint, etc.

So the cleverest will work just like our grandparents with those developing trays: they will extract some areas in the picture in some way, will dither the extractions and then correct the local balance of colors through them. So the final step is to understand what the respected grandson has done indeed.

He has extracted some average color, even that greenery for example, and made the general shift of colors in this area. In other words he has added the blurred corrections to the sharp image. Right, the yellow dandelions have floated along with the greenery, although the colored contrast was saved completely. What’s left is to arrange it in the formal procedure.

Let’s take the usual picture of the usual village life taken on the negative (pic 1 and 2) where the levels and balance on grey are already corrected. Firstly, get rid of the blurring of the colored component (blur the copy and lay it over the original picture in the Color mode, or switch temporarily to the Lab and blur the colored channels) in order to close this question for ever. 

1

2

On the 3 picture we see that the flowers became more green, lips are grey and the teeth are pink (vamp). Yeah, we see that. Even before the correction we already have some drawbacks. So we go back to the original picture and will move only towards right direction.

Duplicate the original picture (Background) and use the Gaussian Blur in order to smooth the grain slightly (pic 4) but not much.

3

4

And again duplicate the blurred layer. And start the corrections; don’t be confused that the picture is blurred.

You can use the entire range of the tools. Just remember that Selective Color works only when you are in RGB. In this case for simplicity sake use Replace Color (Image l Adjustments l Replace Color…) inserted it in Action several times one after another. Firstly, extract and overshadow the greenery (pic 5), then highlight the flesh-colored tones (pic 6), and blue reflections from the sky and under the eyes (pic 7). In this case we had to change not just the brightness but the tint also.

5

6

7

Now we have two blurred colors: with original colors and brightness (Blurred Color layer) and with changed ones (Blurred Color’ Corrected layer). We got their differences Delta (L, a, b). Rename the corrected layer and (already in Lab) let’s subtract from all its channels L, a, b the correspondent channels of the blurred layer with the original colors (pic 8-10). And now we have such a difference girl (pic 11). And blur her image slightly again (R – 0,5-0,7 px) as groups of colors in the Replace Color tool are extracted rather roughly, linearly (pic 12).

8

9

10

11

12

The next step is to remove the unnecessary blurred layer with the original colors, extract the copy of the original picture created beforehand (a sharp copy) and add to its sharp channels L, a, b the correspondent blurred just received differences. The channels state before and after combination is shown at pic 13 and 14; and on the 15 pic – the result for one of the colored channels (b). Don’t get confused that everything is vice versa at the screenshot of the dialog box: Subtract instead of Add, Invert mark,  and offset= + 127instead of -127. It’s just Photoshop 8.0 fails if the offset is less than 0 at 16 bit mode… By the way, you could also use blend mode - Linear Light but with Opacity = 50 %.
 

13

14

15

We are almost done. Let’s compare the results. During each using of Replace Color I have written down all its characteristics and repeated the procedure with 0 blurring (it’s an equivalent to the case when the tool is used directly to the exact original picture) without any complicated moves in Lab and out. The direct using of Replace Color has lead to tragic results with the greenery, girl’s face and especially under the boy’s eye. And the using of blurred differences applied to the exact original picture provided the natural effect. Compare the pictures 16-18 where the original picture, direct using of Replace Color and the method of blurred differences are shown.

16

17

18

Have a look at the picture where I have additionally darkened the sky and made it more intense and played with the sand a bit (pic 20, original is pic 19).

Also I have tried to change the dark-lilac shirt of the gentleman with a bright-green in the same method. I just wanted to know what would happen. But everything has its limits and the spirit of Photoshop has kept me softly from doing it…

19

20

Of course no one has cancelled the correction efficiency of different areas of the image through the blurred masks. Moreover, the most accurate results, for more time though, can be received in this way. And there is something to be improved here. For example to make the convenient tool Color Range react only for color but not for the brightness. One can refuse from it and create more harmonic mask by using some manipulations at Lab with the selected color and the blend mode – Difference (although you can not make some additions to the colors then). Besides, the independent masks for the colors and brightness come easily and naturally, which can be combined then as you wish.

Dynamic management of layer-masks

It is possible in Photoshop to manage dynamically the layer-masks and mix them to some degree. For example we have created two its components (bright one and colored) and want to use them without mixing physically. Above the Adjustment Layer which performs the required correction but without mask we place the two layers (unchanged Curves for instance) with the inverted masks of both types and make the Knockout Shallow mode to the layers. The white spots of the masks layers cancel the actions of the corrected layer and its actions are summarized. By duplicating the layers you can narrow the effective mask in this or that component dynamically.