Small correction: vips_tonemap() was misspelled, but is vips_tonelut().

Hello @CoachRDeveloper,

Usually you'd transform to CIELAB, tonemap the L, and transform back to RGB. nip2 will to do this automatically I think. Perhaps:

But of course it depends on your needs. The libvips tonemap operation was originally designed for prepress work.

Generally you'd hire a good photographer and they'll do all this for you in lightroom or the phase one thing.

Many thanks for reply and providing the information. As taking photos is automated and cameras preinstalled, hiring a good photographer is not realistic. Usually it is lack of willingness to invest in enough LED tubes or changes made by the owner of the studio to e.g. the floor (e.g. switching to a black floor) without adjusting the required amount of lights. Therefore, we have to rely on some tricks to enhance the photo quality to some extend.

From what I tried myself (looks much like you did with tonelut), I noticed that the red color became less vibrant when enhancing the shadows.

But which operation did you apply to enhance the photo? I have to do it programmatically in C++.
Is adjusting the brightness not a good way to reduce the flooded black shadows?
Or is HDR something that is useful to address this?

Please forgive me my lack of knowledge about when to apply enhancing brightness, tone curve or HDR (with the use of libvips, as our code already is using libvips).

I forgot something: is there something as 'a generic tone mapping' curve (=LUT table) that can be applied to any car (under same lightness conditions, taken from same location), even if the car color is different (which can be any color)?

And how to address aging of the LED tubes (i.e. reduced light output).?

It is somewhat unclear to me what the difference is between VIPS_INTERPRETATION_LABS and VIPS_INTERPRETATION_LAB. I guess the latter is 8-bit, although not explicitly mentioned on https://www.[libvips](https://www.libvips.org/API/current/VipsImage.html).org/API/current/VipsImage.html, but for VIPS_INTERPRETATION_LABS, I don't need to set in-max/in-min, as the default (short int values) seem to suffice.

I made a tiny demo prog for you:

/* compile with
 *  g++ -g -Wall tonemap.cpp `pkg-config vips-cpp --cflags --libs`
 */

#include <vips/vips8>

using vips::VImage;

int
main(int argc, char **argv)
{
    if (VIPS_INIT("")) 
        vips_error_exit(nullptr);

    auto image = VImage::new_from_file(argv[1]);

    // import using the embedded profile, and import to XYZ PCS
    image = image.icc_import(VImage::option()
        ->set("embedded", TRUE)
        ->set("pcs", "xyz"));

    // brightness boost ... this is best done in XYZ space, where pixel values
    // are proportional to the number of photons
    //
    // 2.5 seems about right for the sample image (bad underexposure)
    image *= 2.5;

    // to labq, then labs .. this route clips out of range values
    // (with 8.16 you can go straight to labs)
    auto labs = image
        .colourspace(VIPS_INTERPRETATION_LABQ)
        .colourspace(VIPS_INTERPRETATION_LABS);
    auto l = labs[0];
    auto ab = labs.extract_band(1, VImage::option()->set("n", 2));

    // L will be ushort with L* scaled to fill 0-32767 
    //
    // take the histogram and find the 1% and 99% points, then set that as the
    // range ... we want to avoid specular highlights and any pools of 0s
    auto black = l.percent(1.0);
    auto white = l.percent(99.0);

    printf("black = %d\n", black);
    printf("white = %d\n", white);

    // make a tone curve for this image
    auto tone = VImage::tonelut(VImage::option()
        // tonal range
        ->set("Lb", 100.0 * black / 32767.0)
        ->set("Lw", 100.0 * white / 32767.0)

        // position of shadow, midtone and highlight in 0-1
        ->set("Ps", 0.15)
        ->set("Pm", 0.5)
        ->set("Ph", 0.8)

        // shadow, midtone and highlight adjust, +/- 30
        //
        // shadows +10 seems to help the tyres
        ->set("S", 10.0)
        ->set("M", 0.0)
        ->set("H", 0.0)
    );

    // use the tonecurve to adjust L*
    l = l.maplut(tone);

    // rebuild the labs image, then back to srgb
    image = l
        .bandjoin(ab)
        .copy(VImage::option()->set("interpretation", VIPS_INTERPRETATION_LABS))
        .colourspace(VIPS_INTERPRETATION_sRGB);

    image.write_to_file(argv[2]);

    return 0;
}

It makes:

Wow, many thanks. And tricks like using LABQ before moving to LABS, I could never have invented this myself. I will try your code today. Again, thanks for great help.

The color temperature is typically determined by the camera, when adjusting the white balance upon first shoot in the car studio.
 
I hope I don't bother you with some other questions:

Q1:
Will increasing the brightness only (the L channel of LABS) with e.g. a fixed value result in a similar result, or will playing with tone mapping provide better results? I do not know which technique saves most details in a (JPEG) image (our source is always JPEG, straight out of the camera). I'm a C++ programmer, not a photographer or designer, sorry....

Q2:
With
vips::VImage vipsImage = vips::VImage::new_from_file(flInput.c_str());
Then
const auto stats = vipsImage.stats();
Returns the statistics, but then rows are the y part I guess? in:
auto agvAll = stats(4, 0);

The documentation https://www.[libvips](https://www.libvips.org/API/current/libvips-arithmetic.html#vips-stats).org/API/current/libvips-arithmetic.html#vips-stats says that row 0 is for all bands, and then 1 and so on per band. However, in the getter https://www.libvips.org/API/8.11/cpp/classVImage.html, search for operator(), then it is not described wether x points to rows or columns. It seems to be rows. Can you please confirm?

I have checked this, because the following lines return the same value, which is hopefully a proof of the previous conclusion:

auto agvAll = stats(4, 0);
auto avg = vipsImage.avg();

Q3: Is there a possibility to let show/trace NIP2 the executed VIPS commands? I guess that vips.exe is being called and with tracing, I would have more information to deduce the required C++ actions (like you now did for me, but I don't like to bother you each time)....

Q4: I just discovered that 8.16 is not yet released. I'm at 8.15.1 (just one version behind last released one), on Windows. When will 8.16 get released?

I know the problem is underexposure. But the owners of the car studio have their own reasons for not upgrading their studio with additional LED tubes. That's why we have to apply some 'tricks' to provide better quality, as underexposed photos are not an advertisement for us. So we have to do something, but still need to convince the studio owner that uprading the exposure conditions are the better approach (which is not easy).

I've tried out your suggestions and the result is very good. The biggest enhancement is in adjusting the brightness and tonemapping enhances the details in the tyres. Thanks again.

Underexposure is best fixed by multiplying by a constant in XYZ space.

Underexposure is a lack of photons (obviously), XYZ space has numeric values proportional to the number of photons, so just scaling up is a very easy fix.

You generally want to get the brightest diffuse reflector in the scene (the left-hand back wall in this case) up to about Y 98. The specular highlights on the car (eg. paint, chrome) can go quite a bit higher than this (they reflect the lights directly back into the camera), so you need to be a little careful what you set exposure off.

If they are happy with the colour balance, then of course leave it, but you would generally aim for the media white point, which is 250, 250, 250 for JPEG, and they are some way off that.