9

This is a photo I'm trying to update to modern standard:

enter image description here

What I did was go to Levels, and for each color channel I moved the slider from a spot with no information to where the histogram begins. I then used selective color to lower the red in the midtones, then I attempted to correct a newly emerging blueish color by lowering it in curves in the blue channel and color balancing it to be less blue as well.

Please let me know what other steps I can take to make the photo look more realistic and with better color balance.

This is where it is currently:

enter image description here

6
  • Hi. Could you please explain what do you mean by 'modern standard'?
    – Vikas
    Jul 25 at 17:14
  • @Vikas basically, make it as color accurate as possible.
    – Alex B.
    Jul 25 at 17:43
  • I tried but I couldn't make it better than what you did lol
    – Vikas
    Jul 25 at 17:45
  • are you using Photoshop?
    – Vikas
    Jul 25 at 18:42
  • @Vikas yep that's the one I'm learning everything on
    – Alex B.
    Jul 25 at 19:21
11

I don't think there currently is a way to get a lot closer than what you did in photoshop without spending days on this.

As you see below my attempt (edited from the original red source img) is not a whole lot better either... I guess in terms of colour correction, in your version the light tones (faces, left curtains, white dress) are a bit too red. Midtones are pretty good. And dark tones as for example suit shadows look to red too.

enter image description here

Afterwards, the main issue I think is that on old film light colours like faces appear too light and loose a lot of colour. As they are pretty over exposed, it probably needs a layer with colour set to multiply for faces. On the other hand, the dark tones tend to come out too colourful (as you can see well on the red curtains in your corrected version).

Probably to start with a more high-res scan of the image would help as well to preserve as much detail on the light and dark tones...

And removing cracks / dirt / noise may help it look less retro, too.

enter image description here

There is also a new feature in photoshop "neural filter" that could work well for this, but that would take a lot of time to do.

5
  • Hi Julian thank you for the response. Could you explain a bit further about the multiply technique? Is this just copying the image and setting to multiply, or should I attempt to 'paint' over the faces to tone down the exposure?
    – Alex B.
    Jul 25 at 23:31
  • The way i've done it in this example is making an empty layer and set it to multiply. then yes paint with a blurred brush – probably best to sample colour from image. Lightning is a bit different for all faces, so maybe colour needs to vary a bit. (of course if you want to do it perfectly it would need to be kind of painted with different colours for shadows and highlights etc.) Then play with the opacity of the layer. Jul 25 at 23:46
  • If it is a light adjustment like the one I've done the paint strokes don't need to be very accurate, as multiplying a light beige on the black hair for example wont be visible anyhow as long as the brush isn't set to hard edges... Jul 25 at 23:49
  • 1
    A bit more work could get rid of those horizontal creases across the middle. Depends on how much effort you want to put into it. As for "looking less retro", well, the hairstyles are going to look retro no matter what, so there's only so much you can (or maybe should) do... Jul 26 at 15:51
  • @JulianSteinmann I suggest your attempt is a whole lot better. At least, it is on my screen which just might be part of the problem. How a pic appears depends as much on the program, the system, the screen and how they interact with each other as anything else… Jul 26 at 20:39
7

I would say there's no best way because it will depend on your requirements too. There are many ways, many software you can use. It's about trying different things. I don't have much experience about this kind of editing, but I'm suggesting a few improvements which you can try in addition to what you did:

  1. Use the Levels layer somewhat like you did, to get rid of excess red tone

    enter image description here

  2. Create two new Levels layers, and in one of them adjust Green channel to bring some green colors and in the other layer adjust Red channel to bring some red. Basically you want to do it for those red flowers, red strip and green stuff. Use a soft brush to mask all area except the flowers and green elements.

    enter image description here

  3. Use Selective Color, Vibrance adjustment layers in case you need more red and green or blue colors. Use Exposure adjustment layer and gamma correction to brighten the bottom half of photo, a bit.

    enter image description here

  4. Finally remove some excess red in clothes or bodies/face and background using Hue/Saturation and masking. Here's the final result and Layers Panel:

    enter image description here

    Layers Panel:

    enter image description here

Conclusion: This is what I like. So it will also depend on what you like. Experiment with different adjustment layers and see what works best for you.

3
  • Thank you for the thorough response Vikas. What is the main benefit of making two separate level layers for red and green, rather than adjusting both colors on the same layer?
    – Alex B.
    Jul 25 at 23:32
  • 3
    Because it is for different areas of the img and therefore it needs 2 layers with different masks Jul 25 at 23:53
  • @Alex B. So that green and red changes are independent and you have more control.
    – Vikas
    Jul 26 at 3:17
7

There's no "best way to colour correct". There are many ways to do it. Colour correction is subjective. Here are a couple of semi-automatic methods just to add to the other answers, because who doesn't like a bit of automation!

  1. Open in Photoshop - do Image > Auto Tone

enter image description here

Auto tone is pretty good most of the time, and will remove odd colour casts.

If you think it's a little too blueish, do Image > Adjustments > Colour balance and add some magenta and yellow back into the mix.

enter image description here

  1. Another alternative is Image > Auto Colour. The result is subtly different.

enter image description here

Further refinement is also possible in the Adobe Camera RAW filter, if you convert the layer to a smart object first. It's nice if you want to bring the highlights/exposure down a touch and increase the shadows a touch, or adjust the colour temperature.

enter image description here

2
  • 1
    I like the final product you did. I did a similar method at some stage. For better or worse, I always fiddle with the Camera RAW filter after every edit to see if it can 'spice' things up.
    – Alex B.
    Jul 25 at 23:24
  • @AlexB - yeah I like the Adobe Camera RAW filter too, it's a bit like having Lightroom inside Photoshop. It's fun!
    – Billy Kerr
    Jul 25 at 23:26
3

My take on improving the image

As you've noted yourself, the main problem with the image is the overexposure. But fortunately it's analogue overexposure, i.e. though the faces have lost most contrast the information isn't completely flattened to a single brightness value, but merely compressed strongly. As such, this can be at least somewhat repaired. The easiest tool is with colour curves. I'll explain the details in a moment.

However, if you try that, you'll run into another problem, namely that after the overexposure the image has received a lot of noise/quantisation. If you were to correct out the overexposure directly, you would drastically boost the noise in those region. Therefore the steps I used are

  1. Scale the image up to 200% (2106×2546), using Lanczos resampling. This prevents any digital artifacts we're going to introduce from taking large blocky pixel appearance.
    (Mathematically speaking, it prevents aliasing in the nonlinear processing steps.)

  2. Apply a selective Gaussian filter. Only a really gentle one (I've used blur radius 6 pixels, with max-Δ of 7). At this point, the effect of the filter is basically invisible, because any features visible in the original will be above the max-Δ threshold (which disables the filter). It does, however, remove noise in the near-constant overexposed parts, which will be important in the next step.

  3. Repair the overexposure. Overexposure is soft clipping, i.e. a sigmoidal transfer function that has only a very shallow derivative at high values. To undo this, you need to approximate the inverse of that transfer function, i.e. a function that has a very steep derivative at those values. It's necessary to do it for the colour channels separately. Example for green (which is the most critical):
    Colour curve for the green channel
    To see where to put the steep segment, look at the histogram. The L-shaped section at the right is the overexposed part.
    In addition to addressing the overexposure, the curves tool also removes the constant offset, in particular in the red channel.

  4. After the main, channel-wise curves, I used another curve in the value channel (i.e. in all colour channels), mostly to light up the scene a little. Though this shouldn't be overdone, else you're back to overexposure.

  5. Scale the image back down to its original resolution. This is, for one thing, just honest (any higher-resolution information is artificial at this point), but it also avoids that the faces look all too much like they're painted on zooming in (an artifact of the two nonlinear exposure transfers), because you just can't zoom in as much.

1
  • I'm trying to unpack this information - most of it is quite new to me as I don't have any training in mathematics. What I understand is that color brightness information is reserved, so I have to recreate the contrast, without directly brightening areas to avoid additional image distortion? So you upscaled the image to temporarily introduce additional information, then smoothed everything out with the Gaussian filter. Then you changed the RGB curve, and then each color channel to create a smoother Sigmoidal curve. Then finally downscaled image size to hide the artificial information?
    – Alex B.
    Jul 27 at 18:32
2

In my opinion the best method of color correcting, which gives you the most control is using color-curves like with this tool in GIMP (I bet there is an equivalent in Photoshop too):

GIMP Color-Curves Tool

It probably has a bit of a learning curve, but you get a feel for it after some time.

The way it works is, the x-axis is the input and y the output. Left and bottom is black/zero and top and left white/full. For every pixel which has the value x it sets the output to where the curve is at on the y-axis.

This means you can darken and brighten any range, from shadows to highlights, each channel individually or combined.

You can also use it multiple times in a row to fine-tune the results.

In the screenshot above, for example

  • I've drastically reduced the red channel while keeping some in the highlights (faces)
  • taken out some blue only in the shadows
  • also removed some green evenly overall
  • changed the contrast a bit with the combined black curve

Here is the rough result after a little extra tweaking: Result

1

You can try online AI Image Colorizers. Some of the free image colorizers I found online are

  1. DeepAI Image Colorization
  2. imagecolorizer.com
  3. colorize.cc

You can adjust the Colorization Factor accordingly. I have included Colorization Factor of 12 and 25 respectively as an example.

image_edited

1
  • I've tried just about every colorization option there is and they are all awful to say the least. The only decent one is the new Photoshop neural filter, which pretty much nails the natural skin tone look. Just need to fix the blotches of color it often reproduces. I could theoretically make this photo black and white, take out the creases, then colorize filter the skin, and use a new layer with color blending to paint in the clothing and background.
    – Alex B.
    Jul 29 at 8:56
1

The method I would recommend is a bit less conventional and may seem counterintuitive, but the results speak for themselves. enter image description here

First, I changed the color mode from RGB to CMYK (I know. Bear with me.) Next, I auto level (Ctrl+Shft+L) each channel in turn individually. Don't auto level all channels at once. Once each channel has been auto leveled, I went back to each channel and fine tuned the levels as needed for a final touch up. In this case, I pulled up the levels for Cyan and drew the mids down to about 1.75, and pulled the yellow highs down to about 233 and the mids up to about .46. These adjustments will vary depending on the photo. This photo still has too much red in the lower half, but that can be tweaked further. Once you're done, you can change back to RGB if you want. I wouldn't normally advise switching to CMYK for editing color, but for these old photos, it seems to work a treat.

Side note: This took less than five minutes.

3
  • 1
    Well, if it works, it works. And you couldn't achieve a similar result using Selective Color?
    – Wolff
    Aug 19 at 19:55
  • I don't think so, since selective color doesn't allow for leveling of each channel, but is just an overall adjustment. When I tried it, I can get close, but it takes a lot longer and the colors end up looking weaker. I'm sure someone with a better eye could do it, but the channel leveling method is dirt simple and very fast.
    – 13ruce
    Aug 19 at 20:09
  • Seems crazy at first, but in this case it's a very clever method using CMYK colour mode. Sep 7 at 0:58

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.