Ceci est une ancienne révision du document !
This series is aimed at learning to make something of the old photos in my possession and others in the public domain, due to their age. You, the reader, are welcome to tag along and, I hope, glean some small insight and perhaps an idea or two from time to time. No promises are made as to the quality of the content, or potential errors and omissions. I am a computer scientist, not a true artist or a professional in image restoration. So please take all this as a best effort, but with no firm guarantees — much as is the case of most open-source software. In the previous part of this series, we took a look at an interior photo which exhibited the typical lack of sensitivity of early digital cameras’ sensor chips. In this part, we will work on the opposite defect: washed-out colors due to too much light in the original scene for the camera to handle. This is typical in outdoor shots where plenty of sunlight has illuminated the scene. Contrary to popular belief, the problem may actually become worse in slightly overcast or hazy conditions, where a lot of reverberation tended to overload early light sensors. This was the case in the following scene from the island of Cheung Chau in Hong Kong.
Cette série a pour but d'apprendre à faire quelque chose des vieilles photos en ma possession, ainsi que d'autres du domaine public du fait de leur âge. Vous, lecteur, êtes bienvenu pour m'accompagner et, j'espère, glaner quelques petites particularités et une idée ou deux de temps à autre. Je ne fais aucune promesse sur la qualité du contenu, ou sur les erreurs et omissions possibles. Je suis un scientifique en informatique, pas un artiste ou un vrai professionnel de la restauration des images. Aussi, merci de considérer ça comme mon meilleur effort, mais sans garanties fermes, comme c'est souvent le cas dans les logiciels Open Source.
There is clearly enough light here to take a clear picture, since even the shadows under the tree branches hold enough detail. However, the sensor was unable to handle the sunlight bouncing off the white buildings in the background, the illuminated part of the tree trunk, or bicycles’ chrome handlebars. But, is this excessive light the case for all primary colors? The Histogram window (at menu Settings > Dockers > Histogram) gives us a more complete story: Going from left to right, in the first place we see that very dark pixels at the far left of the histogram are few, though among these there is a tendency to contain a blue tinge and a defect of cyan. There is, in fact, a large proportion of pixels that have intensities in the middle range, with some reddish tinge to them. There is also a distinct lack of light-colored pixels in the middle of the top quarter to the right of the histogram, and these have a magenta tinge to them. Finally, there is a very heavy group of very light (whitish) pixels to the extreme right of the graphic.
Thus, from a standpoint of light exposure, what we have seen in the histogram shows us that the image is in fact globally rather well-balanced, except for the very light pixels. This tells us more about how the (old) digital camera’s chip reacted to light in the first place. Its response was correct for low and medium levels of incident light, but exaggerated for higher levels, which leads us to deduce a lack of calibration of the software that translated electrical signals into colors levels in the controlling software. On the other hand, there seems to be some relationship between color balance, and pixel intensity value. Let us examine this further. The Layers Docker that is usually found in the lower right-hand corner of Krita’s window has a second tab, Channels. Using this tab, we can activate or deactivate at will any or all of the main Red, Green, Blue and Alpha (transparency) channels in our image, without actually making any changes to pixel values.
After some tests, it becomes clear that all channels have, in fact, lost a lot of detail in the lighter pixels. Of the three main colors, Red is the channel that holds more information of the darker colors – and, thus, it is the channel that represents better the details in the shady part of the tree. In other words, the chip inside the camera seems to have been more sensitive to red light than blue or green, especially at higher intensities. At this point, we can take our handling of the image in several different directions. The first would be to try to balance intensities on a channel-by-channel basis. In Filter, Adjust and Color-adjustment curves, I started by giving the general luminosity response a slight tweak in the middle of the higher quarter of the scale. What I am doing is counteracting the corresponding peak at the rightmost end of the histogram, and trying to eek out some more detail from the washed-out whites. This is clearly not working.
A very slight increase in the details of buildings at the back of the image is achieved, but at the expense of rather weird-looking colors, for instance in the greenish pane windows to the left. Could I get better results by working on a single channel? As before, I went into the Color adjustment curves window, and now tried adjusting each channel individually. The end result is the very same: details are not forthcoming. It is in fact clear that a large quantity of information has simply been lost inside the lighter parts of the image. Whichever way we go about it, once lost, this information cannot be retrieved since it is no longer within the image. So, what can be done? An alternative approach is to step back and think about what we have been doing. The main center of our interest in this image is clearly the tree itself, and the heap of bicycles parked around it. The buildings in the background are less noteworthy, and give very little to the complete message. So, what if we exaggerated the process, and blanked out the burned whitish buildings altogether? We could thus concentrate on the photo’s main subject, with fewer distractions.
We can do this either in color, or in black-and-white. When we examined the image’s color channels previously, the red channel actually had quite a nice vibe to it. So I went back into the channels part of the Layer docker, and turned off both the blue and green channels. I then went back into Layers, and added a new transparency layer. Then, using the airbrush at low opacity (about 30 to 50%) and some pure white, I scrubbed out selectively parts of the buildings in the background, concentrating on their darker elements that stood out more such as window frames. Other vegetation such as some palm trees in the background were also cleared up. The end result allows the main subject to stand out from its surroundings, and the eye can concentrate on the details that struck the photographer’s imagination when taking the photo more than a decade back.
In this part of the series, we examined the effects of excessive light on one of the first digital cameras, seeing how very light colors got easily washed out. Since much information on details has not been retained within the image file, it is now impossible or very difficult to put it back into the image through color curve manipulation. Thus, it may be advantageous to proceed otherwise, thinking more about the global effect that is required and trying to use the burned-off parts of the original photo to our advantage. In the next part of the series, we will work on another early digital photo, in which the camera’s optics and light-sensitive chip lost some definition in an otherwise quite pleasing scene, due to their lack of resolution at the time. Until then, take care!