Refactor and clarify the specification of Image containers

What are the problems?

Colour image

There is currently no clear specification about what shall contain a ColorImage. Is it sRGB, is it RGB?

Why is this an issue?

The colour image have several save function which more or less write the raw colour value to the file regardless of the format. But, a EXR is generally a linear RGB while a PNG is sRGB if no ICC profile is provided.

Hence, doing

colorImage.save("img.exr");
colorImage.save("img.png");

does not have a very clear behaviour.

Furthermore, savePNG multiply the values by 255... So, it is already applying a conversion.

What do I suggest?

A - ColorImage shall be an abstract class (will develop on that later)

B - ColorImage shall assume the pixels are linear values (XYZ, RGB ? I would be for XYZ)

C - ColorImage shall be renamed accordingly (XYZImage or RGBImage)

Refactoring

I suggest a refactoring of the hierarchy of the Image classes. I am not an UML expert so excuse the approximations but here it is.

Each of the Image shall implement the copy operator. This will ease image conversion.

Also, I would avoid public load instance functions. I would prefer constructor in the corresponding class (for instance SpectralImage *img = new EnviSpectralImage("foo.hdr");). That does not mean we cannot have static load factories in each class.

Since this architecture is highly a personal taste and not the only way to implement it, I am opening discussion (and probably trolls ;) ) threw this issue.

Questions about colour conversion / tonemapping

I have the feeling that we do it wrong for Reinhard tonemapping.

I think we shall consider this tonemapping function (and any tomemapping function) transforming a linear RGB space into another RGB space represented as linear value. What do I mean? I think we shall apply anyway a sRGB gamma correction whatever the tonemapping function we are using in case we intend to display the result on a 8bit / channel monitor or on a image without an ICC profile. If there is an ICC profile, the colours shall be consider as linear RGB and then converted in the specified ICC colourspace.

Why? A tonemapping function is "just" a luminance / colour adaptation. sRGB is an old standard still in used to colour correct an RGB colour on a non linear monitor (cathodic monitor). This have nothing to do with the perceptive tonemapping function.

Am I correct there or do I miss something?