PC Plus HelpDesk - issue 216

This is our original photograph of the Big Ben tower in London.

It has good highlights and shadows and poesesses red, green and blue so we can see what is going on when we start messing around with it.

If we just make a straight negative, we see that blacks have become whites, reds have become cyans (bus) the blue-cyan sky has become brown (our eyes are better able to distinguish between reds, oranges, yellows and greens than cyans and blues - this is why there are so many words that describe them and so many people think that cyan is blue) and so on. To return the colours to their individual values (ie, red back to red and so on), we have to rotate the hues by 180 degrees.

Instead of doing that, we could just separate the image in to HSL channels (Hue, Saturation and Luminosity) and then, make a negative out of the luminosity before recombining them back into the composite - ie turning the HSL back into an RGB image.

Negative with colours restored by rotation

Both of these methods produce the image on the right which still has a blue sky, red bus and yellowy-green leaves on the trees but instead of a near-white sky at the horizon, it has turned close to black.

Separating the Luminosity out and processing it on its own (thus saving having to reprocess the hue later on in order to recover the colours of the picture) has the advantage that we can mess around with it in any way we like. On the right, I have solarized (or more correctly when talking about image processing on the computer, pseudo-solarized) it so that the highlights become inverted and the shadows stay the same - it is just as easy to do the opposite, either by inverting the pseudo-solarization transformation graph or by inverting the luminosity image after it has been pseudo-solarized.

Pseudo-solarized image with retained colours

On the right, we have the final composite image with the highlights, mid-tones and shadows pseudo-solarized as you would expect but the hue information in the colours in both the normal and inverted portions of the image has been retained.

Pseudo-solarization best works where there are bright skies and intricate structures that protrude into them. Whilst a lamp post is not particularly intricate, it does demonstrate that this can work even when the intrusion is only a pixel or two across

On the right is the MI6 building.

So, we make a copy of this and then turn that into a greyscale image so that the luminosity information makes sense (remember that greens are brighter than blues and so on).

Thresholded version of greyscale image - making a mask for future processing

Next, use the threshold function so that we create a black and white mask. Here, it is done at 50 per cent but you can use other values if you are prepared to sort them out later on (you could adjust the luminosity mid-point of the image to start with so that the 50 per cent threshold separates the image the way you want it and then change it back later).

Masked image (threshold image in alpha channel) and then negative

Next, go back to your original colour image and select the alpha channel (this should be all black as you haven't got anything masked off yet). Then, paste the threshold image you generated into the alpha channel and, if necessary for your image processor, combine it with the alpha base.

With the mask in place, you can get out of the alpha channel - you will see that the mask outline matches your threshold image - and then invert the masked part of the image so that the brighter part of the image is now effectively pseudo-solarized.

masked part of image with colours rotated

Next, and with the mask still there, rotate the hues by 180 degrees. In the picture on the right, the sky has become blue again.

Next, remove the mask and then adjust the tone balance so that it covers the whole range, brightening up your image.

either: negative and colour-rotated version of previous image or; use inverted threshold mask.

If you wanted the sky the other way around, either invert your threshold mask or take the image you have just created, invert it and rotate the hues - you get the same result.

The above transformations were performed using the HSL colour model in Micrografx Picture Publisher. However, there are many other colour models and the one in the GIMP is HSV (Hue, Saturation and Value). Whilst it is possible to invert an image and rotate its hue or even solarize by using masks and rotate the inverted hues to get the images above, just taking the Value (which very tenuously replaces the Luminosity) and inverting that in the same way that you would with the Luminosity in the HSL model simply does not work. The images on the right are the result of inverting one or some of the HSV images and then recombining and whilst interesting in their own right (and not easily reproducible using HSL), demonstrate that you need the HSL model to do the above properly.