Table of Contents Show
Archive notice: This article was originally published on May 1, 2017. Links and embedded videos are preserved as part of the historical record.
Any good software should solve a problem (and not create new ones), but what problem are we talking about here? Every camera professional knows that a camera does not see colours in the same way as a person. We are all familiar with the fact that a correct adjustment of the colour temperature is necessary so that neutral tones such as white or grey are displayed without a clear colour cast. Human perception has no problems at all with perceiving colours as “correct” under strongly changing lighting conditions if we are exposed to lighting with the same colour spectrum with our entire field of vision.
However, the problem goes even deeper, because no camera can correctly reproduce the different sensitivity curves of the human eye with filters that separate into the primary colours red, green and blue. We also separate three different colour ranges, unlike some animals: The mantis shrimp has eight different receptor types, whereas many mammals only have two. However, the spectral curves of our eyes differ considerably from any camera that can be realised at a reasonable cost and can only be reproduced imperfectly with three colour filters and silicon chips, which only measure the light intensity. In addition, cameras with only one chip have the challenging task of de-bayering, which is made more difficult by the secondary densities and side light incidence of the less perfect, microscopically small colour filters.
Each manufacturer makes its own stylistic decisions as to which compromises are made when translating the measured values via the colour matrix in the camera. Many also offer a number of options for tinkering with the “look” in the camera itself. In addition, different lenses can influence the colours. The task of colour graders is then on the one hand to coordinate the look of different cameras, but primarily to give viewers the impression that they are seeing a real situation as if they were present in the scene. This usually requires a great deal of experience and empathy.
Colour consistency
In evolutionary terms, the eye is originally an outgrowth of the brain, and in fact our retina is already a kind of preprocessor for visual data, which is processed by the brain in a very complex way. Even if we can adjust to changing spectral distributions to an astonishing degree (unlike a camera without white balance), we still have certain expectations of the colours of certain objects.
It is well known that viewers can recognise quite accurately when something is wrong with human skin tones – after all, we constantly have the opportunity to compare them with the original. But important colours from nature such as the blue of the sky and the green of plants are also perceived quite critically and should be portrayed credibly. But in the blue to green range, cameras (especially if they offer good skin tones) often have problems, as we have repeatedly found in our camera tests using a colour chart and vectorscope.
The developer approached the problem from the side of human perception and sought to develop an algorithm (called Xe8472) that mimics the processes involved in human colour perception and does not merely attempt to compensate for the specific weaknesses of individual cameras. He has been developing the process since 2001 and first published it in 2002 for correcting photos. By his own admission, however, it took him over 14 years of further research and development to optimise it for film purposes. Anyone who has ever tried to process a film with HDR software for photos knows the challenges of the moving image: the small fluctuations in the technical parameters between the individual images can lead to irregular flickering in automated corrections. In addition, a photo is perceived as a singular event, while viewers can critically compare the colours in the film with every cut.
Similar to good denoisers, the plug-in therefore analyses a total of three images in the vicinity of the individual film frame in order to achieve stable results when calculating its colour and contrast correction. Incidentally, this is calculated using the more computationally complex method of tetrahedral interpolation (as opposed to the usual trilinear method). Only Resolve offers this method as an option in popular software. It avoids banding and deviations from the grey curve, which (in addition to the weaknesses of 8-bit colour depth or compression) can also occur with the usual calculation of LUTs. Nevertheless, the software runs surprisingly fast with modern graphics cards because the calculation is largely performed by the GPU.
The results
The effect on film images with correct white balance from the camera is subtle at first glance, as good grading should be. But a trained eye will recognise the changes immediately: you almost get the impression that a veil has been wiped from the image. The result is a freshness and naturalness that is particularly impressive in landscape shots, but skin tones from good cameras also benefit in a subtle way. The filter therefore appears to be primarily suitable for documentary films, where you generally can’t afford any elaborate grading anyway. The developer advised us not to even try to understand how it works.
I will therefore limit myself here to subjective impressions: The software enhances contrasts and colour saturation to a certain extent, but is by no means as brutal as the usual automatic process used in editing programs, which simply sets the brightest value in the image to white and the darkest to black. The new process, on the other hand, preserves the intended or naturally existing mood. It also seems to work to a certain extent with simultaneous colour contrasts, as largely monochrome material with lots of green or brown is less intensified than multi-coloured motifs. The website www.EPICOLOR.film contains convincing film examples, including the reconstruction of historical colour films.
The plug-in can also be switched from source material in Rec. 709 to log, so that the use of a camera with log recording can also be worthwhile where time and money did not previously allow manual grading. This works very well with many cameras, such as the Alexa or the cameras from Blackmagic. However, every manufacturer handles log formats differently, so that some still look too low in contrast or too bright when using automatic grading alone. This can be easily remedied by using the manual controls, as corrections to “Automatic Contrast” (default 85 for log), “Brightness” (a non-linear gamma curve) and, in the worst case, “Contrast Overdrive” (a classic S-curve). Such corrections can be saved as presets so that cameras with different log formats can be matched quite quickly.
Every good colour grader will first want to create a neutral grade and create consistency within the scene, no matter how expressive the look. In this respect, it makes perfect sense to use this plug-in for the initial correction and then build a specific mood on it. The tool already offers two moods with “friendly” as a warmer and “understate” as a cooler variant in the automatic range – basically just an indication that you can create your own presets. A few crisp Hollywood-style looks are also included, and you can load your own LUTs in .cube format.
A common form of automatic colour matching is the “Match” function in Resolve, but this requires a suitable colour card to be recorded. As this function is very dependent on the brightness of the source material, we compared the result with a version from the new plug-in that had been adjusted accordingly in terms of contrast and brightness. It is noticeable that it shows less saturated colours and remains brighter as a result. If you transfer the values obtained using the colour chart in Resolve to a test scene, the result also appears darker, more saturated and slightly more yellow. On the other hand, you don’t always get a shot of a colour chart from the camera people and it can be seen that the result with the new tool may be technically less correct, but also looks very pleasing in direct comparison to the real scene. Only the background is a little too bright compared to the subjective perception, but ultimately a very high contrast had to be converted into the tight range of Rec. 709 for this motif – there remains a considerable degree of subjectivity.
Super white?
The fact that many external filters in FCP-X regularly only use the range permitted for broadcast (i.e. Rec. 709) is problematic. If you have a camera that can use the so-called “Superwhite” range with values up to 109 IRE (e.g. various Sonys with the Cine-1, -3 and -4 profiles), it is mercilessly cut off at 100 IRE. This is not only the case with this software: With Color Finale, for example, you would have to work with a complicated workaround. You can remedy this by activating the “Colour correction” effect and its “Exposure” function. If you pull down the highlights there, the additional details become available again. Unfortunately, the influence of Apple’s filter extends to the centre areas, which adversely affects large areas of the image information for further processing.
The “Levels and Curves” filter by Graeme Nattress (also at FxFactory) does a better job, where you can access the “Superwhite” and adjust the correction curve individually. According to the author of EPICOLOR, he is working on a solution to tackle this problem fully automatically using an algorithm. The software currently runs under FxFactory and therefore only on the Mac, namely with FCP-X and Motion. Resolve could handle the values outside the broadcast standard more flexibly, and the developer already has a corresponding version in the works. It will be available free of charge to purchasers of the current licence after completion. But even now, there’s nothing to stop you from using it to process a film, extrapolate it in a high-quality format for other platforms and transfer the cut via .xml.
Automatic & Manual
Of course, even such an “intelligent automatic algorithm” is not clairvoyant, so that weaknesses of specific cameras or rough colour casts cannot always be eliminated in automatic mode. The examples shown by the manufacturer on YouTube prove that the software can remove colour casts from old analogue film with astonishing effect. This is not always possible in the same way with video material using only automatic mode. If, for example, the white balance of the camera was set to automatic, dominant colours in the image (such as the green in this case) create colour casts in the opposite colour, so that our street has a magenta cast. The tool does not remove this fully automatically, but if you go to the manual settings and use “Colour Mood” and “Colour Mood Intensity” to compensate for the shift.
The software does not fully compensate for a grossly incorrect colour temperature setting, such as daylight instead of artificial light. However, this is not necessarily desirable, as warm-toned material from the golden hour or cool material from a cloudy, rainy day should generally retain its lighting mood. Shifts caused by incorrect Kelvin values are already largely eliminated with many cameras, and if necessary you can quickly compensate further by selecting “friendly” if the tone is too cool or “understated” if it is too warm. If you want to get it even better, manual adjustments or your own presets can also help here.
The plug-in even offers a secondary colour correction, called “Selective Color” here. We used this in the example to get to grips with a massive miscolouration caused by extreme blue, which clearly overtaxed the gamut of a Sony A7R II. Similar action would have to be taken with blue LEDs that exceed the gamut of many cameras (apart from the new IPP2 in RED). It is more difficult to control the tendency towards a green tint with Slog-2 from a semi-professional camera in automatic mode. However, most colourists would struggle with this if they tried without a specific upstream LUT. If, on the other hand, the extremely soft Slog-3 is used with professional equipment such as the F55, such problems do not occur. A respectable result can quickly be achieved by manually enhancing the contrasts and colours.
The very soft Slog-3 of a Sony F55 just needs a little coaching.
Comment
In the already largely overcrowded field of looks and LUTs, EPICOLOR takes a completely new, perception-based approach. For an “epic” feature film, stylised looks are more likely to be used, but the new tool makes colours look amazingly fresh, natural and vibrant. However, I wouldn’t go so far as to call it “completely automatic”, as problematic camera profiles or specific colour errors require manual intervention. But then even log recordings can be prepared in a presentable way without much effort.
This plug-in is a great help for film projects that lack the time and money for time-consuming manual colour grading. But even experienced colourists will appreciate it if no standardised colour chart has been recorded and you want to quickly obtain neutral source material for your own look.
