Merging for HDR :: On1 Photo RAW 2018.5 .vs. AuroraHDR 2019

The photos on the track also gave me a some clearly high dynamic range scenes to compare the two HDR merging applications I currently like.  I began with a set of 5 bracketed photos (-1.3EV, -0.7EV, 0.0EV, +0.7EV,+ 1.3EV) and used the RAW versions on these (shown below is the unaltered jpegs straight out of the camera)

I then merged into both packages using the align image and de-ghosting options in both systems. The On1 version being faster to get to the basic higher dynamic range image (with a generic tone map) which you can pick that up and do further tone and colour refinement as desired. As noted previously the new version of aurora actually refines the individual input brackets using an AI guided system (called by Skylum their Quantum HDR engine) , which takes longer to get to the default tone mapped image. However that default tone map is somewhat optimized to the features of the underlying image. The results where just cropped, but in the case of the AuroaHDR image also vertically transformed to avoid the converging of the tree tops.

The results are both fine, but which is better? The photo on the left comes from On1 Photo RAW and it definitely has done a magnificent job with the shadow detail however it is a little flat compared with the Aurora HDR on the right, which has stronger contrast and slightly more vivid colour. I’m sure I could easily further adjust the On1 Photo RAW output to also show stronger contrast and colour but for now I’m happy with both but Aurora being the new toy has become my favourite. Of course Aurora HDR can not stitch panoramas.

It is worth mentioning that my version of Aurora has been recently updated to version AuroraHDR 2019, whereas my version of ON1 Photo RAW is still version 2018.5

The Big Vert-a-rama

Stands of Gum trees can be hard to photograph because they are so tall. Also I was looking for examples of this isssue of showing verticality for my up coming sketch crawl (in the context of David Hockney’s observations on photography and seeing for his bigger picture series).

In front of me on the track I was meandering along (actually doing detailed sketches, like these on the right). I looked up and in the strong Australian afternoon light was typical challenging composition, the understory in shadow, very tall straight trunks, some almost white in full sun others dark in shadow but the canopy just lacy against the deep blue sky. I know a single exposure is a recipe for blown out sky and/or blown out shadows. So I grab my little Olympus switched over to multi-bracketed exposure. I’ve been really pushing hard against Aurora HDR 2019 and hitting it with 5 exposures (-1.3EV, -0.7EV,  O.OEV, +0.7EV, +1.3EV), to see if their was notable differences in the image quality (not so noticeable in detail but possibly there is in better colour rendition) Then I did three “compositions” the path and understory the top of the understory and tree trunks and finally up above into the canopy.  Rather than holding the camera to a given exposure/ISO regime I left it in automatic so the average exposure could suit the section of image being photographed. So I had 15 exposures (saved in both Jpeg & RAW formats ie 30 files). I used the RAW files in Aurora, to start with higher dynamic range.

First I ran each set of bracketed exposures through Aurora HDR 2019 and save the default tone mapped image (ie no tonal tweaking or use of “looks” (presets). So far so good. I saved them as as both .tiff (to keep the extra bit depth of colour) and Jpeg (sRGB) to show here.

Finally I assembled the three tiff files using On1 Photo RAW 2018.5. I did a little rotation and cropping and exported the full image (now 4079by7105 pixel) but no other edits now as a final jpeg (back to sRGB). There is great detail in the understory and also a nice blue (not overblown sky). This combination has worked well, below is a link to the image on flickr which is the full resolution (click on it to view in flickr)

Getting more colours with Bit Depth

Another area where there is a lot of misconception about colour is the topic of colour bit depth. Really there should be, because it is a simple the numbers of binary numbers (0 or 1) have to describe each primary colour (red, green, blue) in the pixel (colour channel) the more unique colours will be available (and the closer the steps between colours will be. It does not necessarily mean that a wider range of colours will be possible (ie colour gamut). This effect is most easily seen in the image histograms.

As an example and 8-bit coding system, of SRGB, gives each pixel up the 8 bits  or 2⁸ = 256 combination. By convention zero (0) is no colour (black for that channel) and 255 is the maximum intensity of colour in that channel. When all the three primary colours are combined there are 2^8*3  = 16,777,216 different colours definable for any given pixel. This is often described as “true colour”. This is often called the “bits per pixel” (bpp)to describe the sum of all three colour channels and that represents a pixel.

An interesting fact is most human eyes can only perceive around 10 million discreet colours, so displaying any image in more than 24 bpp will go unnoticed.

Most modern camera, will capture images in 8 bit SRGB (24 bpp) and standard .jpeg has this bit depth. Some higher end cameras now offer other colour spaces (Adobe RGB or ProPhoto) and greater bit depth. Remember you probably will not be able to see the difference in terms of enriched colour or image quality. The extra bit depth however can be very handy for post processing and particularly when “stretching” the tonal range of an underexposed RAW file (which can often lead to colour banding in the shadows). Much photo editing and computer graphics software can handle 16bit colour and  .tiff formats can be saved up to this bit depth.

The Cambridge in Colour site has a simple tutorial of Bit Depth, including a great visualization of the effect of bit depth. The African Shutha project has a wonder summary of the topic my Graeme Cookson.

PhotoProject :: My #HandDrawnPhoto approach to #AIart

The success of the Belemy portrait auction, has encouraged me to compare Obvious’s DAG approach to my use of Google Deep Dream algorithm. Deep dream uses a convolutional (foreword chaining) neural network to find and/or enhance patterns in an image. This can be applied in two ways Style & Inception (google’s terms). The inception process tries to find the characteristics of a given object/pattern within a photo and tends to produce surreal images. Style on the other hand  seeks to put characteristic patterns back into a photo based on features the trained neural network has recognized. This is the technique I have become very interested in as a way to put my characteristic marks (and sometimes colours) back into a photo. Hence the name Hand Drawn Photos for my approach. The method does have superficial similarities to Onvious’s DAG Approach.

Here I am using my sketch for inktober as the image to be processed and a photo of the same trees as my training image used to build the neural network (AI). The Photo here is acting like the discriminator in Obvious’s approach. The result (below) is tonally similar to the photo and kind of a messy squiggly version of something I might sketch.  But…

In this, my preferred approach, I am using the photo as the image to be modified and my sketch as the image to train the neural network and letting the system modify the photo to “find” my characteristic mark making (aka style) and modify the photo. In this case I am taking the colours from the photograph, but that is not essential.

I feel this second result is much cleaner and fresher, closer to something creative and artistic, Although I am strongly guiding the outcome by using my own sketches (as opposed to an automated algorithm). So the big question is “Do you consider this #HandDrawnPhoto to be legitimate AI art? I’m warming to answering yes.

If you are interested here is the link to my profile and public images on Google Deep Dream website. (Its free to join and use)

Love to hear your views,  please leave your comments here.

The Obvious & the “first” Auction Sale of #AIart

Just a short break in the discussions of digital colour.

I have been following with some interest the much hyped Auction at Christies of a the “first” AI produced artwork to be sold at auction. The expect price range was USD$7,000 to $10,000 but the actual sale price was USD$432,000 (or around AUD$612,000), That’s a lot given the picture is just 70cm by 70cm. Ok its nicely framed, but…! However I’m sure the value has been realised because it is a first and also the technology and artistic intent behind the image.

The image was produced by an AI (artificial intelligent) under a Paris-based collective of artists, calling themselves Obvious. Their object was pretty simple “Can an algorithm be creative?”. The their approach has been to carefully select a dataset of 15,000 portraits (this is a very manual step by the way) done between the 14th and 20th centuries, which showed a number of characteristics they where interested in and used this to train their AI (presumably a neural network). Next they applied a GAN (Generative Adversarial Network) whereby two algorithms compete to find and acceptable outcome. They called there algorithm the generator (which took an input of random noise and using the trained network created a “fake” portrait which the Discriminator would then rate are a real or fake outcome from the combined trained results, In simple terms the generated portrait was accepted if the discriminator was tricked that it was actually the features expected (from the training set). This stakes a lot of data processing and their images where small. So they added and additional step where they upscaled the image (again algorithmically) to imply a higher definition outcome.

In a fitting testament to the method the collective have signed the portrait with the mathematical formula of their discriminator.

If you want you learn more I suggest you start at the articles on medium at the bottom of the Obvious web site. You will also find other portaits obvious have produced of other members of the fictious Belamy family.

They name Belamy being a light hearted acknowledgement of Ian Goodfellow one of the originators of the DAG approach (ie Good fellow loosely in french Bel Amie)

Just How Many Colour Models/Spaces are There?"

The complexity of colour doesn’t stop at RGB versus CMYK or even the traditional colour wheel. Different industries and investigators have established a myriad of others ways to express colour systems. For now it is enough to group them into 5 main groups.

RGB, (Red, Green, Blue) is used in Camera Sensor, LED Computer Screens and Digital TV Screens, they use these three colours from light emitting or sensing technologies to produce any given colours, through Additive Colour Mixing of the light. Its not really a colour space but a colour model.

Unfortunately there are several variants and this is the first place many digital photographers can get caught out. in 1996 HP & Microsoft cooperated on developing a stand 8-bit colour space based on RGB and it was subsequently adopted as standard for monitors, printers and widely internet applications (eg Browser) and is know as sRGB. With the result that most printers, monitors and software can now correctly render this colour space. When in doubt this is the best colour space to use to avoid disappointing changes in colour and tone. In the meantime other colour spaces with greater bit deep have developed such as Adobe RGB and ProPhoto which can in theory render more colours, BUT your monitor or printer may not be able to show them.

CMY[K] (Cyan, Magenta Yellow)The CMK model is relative new, as it required intense and transparent  synthetic inks & dyes the can mix cleanly. Also the technology of Halftoning (or screen) whereby tiny dots of ink are printed in a pattern small enough for humans to perceive a solid colour, A set of separations for each primary colour was made and overprinted with close attention to properly registering the images.

These days there complex but reliable colour space converters that can take an RGB image and render it in the closest CMYK colours and these are usually built into your printer drivers. This approach forms the fundamentals of ink Jet printer technology., Older printers, including some high end larger format printers may still need conversion (or even tone separation) carried out separately. However most photo services will accept SRGB and do conversion  automatically to CMY if required.

LAB (or CIELAB) is a special colour space in that it includes all perceivable colours.   It is extensively used to compare the colour rendering and matching capabilities of a wide range of technologies and devices, particularly the CIE XYZ graph shown on the right. The L is for lightness and the A and B represent Green-Magenta and Blue yellow components but they are non linear mappings with elaborated transformation functions. However the key is the the three axes use real numbers (rather than positive integers, for bit mapped colours) so an infinity number of colours can be represented. The CIE chromaticity diagram, shown on the right, covers all the colours visible to the human eye and the outside of the convex curve enclosing the colour space show the Wave length of light that corresponds with that colour. You are likely to see this diagram when a manufacturer is extolling their virtues, aka wide colour gamut, of their new devices

Adobe’s PhotoShop has a LAB mode to allow device-independent colour.

.

HSL or HSV, are the cylindrical equivalent of the RGB additive colour model but include the brightness of luminance L (sometime B for brightness), as well as hue H as a radial measure and saturation S as distance from the centre. The system was originally invented in 1938 by George Valensi as a method to add colour to and existing monochrome (L signal) Broadcast (see also below how this might be encoded). The V in HSV stand for value and in the variant of the colour model to top of the cylinder is white and the base if black and may better represent how paints are mixed. It is frequently represented as a cone. This model has been widely accepted and applied in most image editing and computer graphic applications and

YUV,  of Y’ (luma) UV (chrominance) is a technology that was widely used in analogue colour TVs , PAL Digital and some movie formats. Its original begin when B&W analogue TV was being upgraded to colour. The Luma channel is exactly the original Black and White signal. The colour channels  U & V utilize the fact that the green sensitivity of the human eye is somewhat overlapped by the red and blue cone receptors and therefore the signals bandwidth could be reduced by not transmitting the green information.  The original TV engineers, following VAlensi model,  brilliantly worked out that rather than use absolute R (red) and B (blue) they could send the U & V the difference from a reference average and tell the TV to just shift the colour of a specific pixel without altering its brightness. Thus an older B&W TV which could not decode the difference signals would just how the normal B&W picture, thus avoiding making older TVs redundant!  When you use the yellow plug, composite video, you will be using some variant of Y’UV. Standard Digital PAL and HDTV also use modern variants of this colour encoding method.

If my very limited description has confused you Cambridge in Colour has a great article of visualizing and comparing colour spaces (well except they use the color spelling)

Now for the really interesting part, many of these colour system describe colours, that we can not see, our cameras (even the expensive ones) cannot differentiate or cannot be reproduced either on our computer monitors phone screens or inkjet printers. In fact most devices have a limited capacity to reproduce colours, and the range of colours they can produce is usually referred to their colour gamut. More on that to come in future posts.

Building a Digital Colour Wheel

Most of us are familiar with the conventional colour wheel. It is the colours of the rainbow wrapped around and joined at the purple/violet segments. It normally shows 12 colours, the primaries (Red, Yellow & Blue) then the secondary colours easily mixed from those (Orange, Green & Purple). Finally the six tertiary colours mixed from the adjacent primary and secondary colours. Whilst many people instinctively know a harmonious colour scheme (eg red & green, Blue & Orange, Yellow and Purple) They might find it difficult to describe why. The tradition colour wheel can come to the rescue here, colours on opposite sides of the circle are called complementary, beside each other are call analogous, In addition if one of the complimentary colours is left out but the analogous colours either side of the missing compliment are present then this is described as a split complementary. All these combinations are know to be harmonious (and desirable for an artist, photographer, home decorator or fashion designer). The colour wheel is very useful and you can read a bit more about Basic Colour Theory at Color Matters.

Some complications arise when you start to look at how cameras and TVs work. They use a three colour palette RGB (Red, Green & Blue). This doesn’t fit onto the traditional colour wheel very well (first image below). These three colours are also know as the additive palette. These three colours do work well in terms of how our eyes (specifically the three types of  “cones” in our retina) recognize colour. A lot of people not familiar with this system are very surprised at the lack of yellow, but if you add red and green light sources you will see yellow. Things get even more strange when you consider the approach developed by traditional colour printers (including your ink jet printer) which use the  CMY [K] (Cyan Magenta & Yellow) subtractive palette. The K in square brackets stands for blacK, as printers find they need a true black, because a mixture of magenta, cyan and yellow inks or dyes tends to be a muddy dark grey and leaves images flat. Black is not really a colour, but I wish to avoid the argument on that for now. It is even harder to fit these colours on a traditional colour wheel (there is no matching segments for Magenta or Cyan for a start). You can read a bit more on the RGB & CMYK colour systems at Colour Matters.

This is the point where I decided to combine these two palettes, and form 6 new primary colours. I’ve since discovered I’m not the first to have attempted this (eg see Warren Mars website) and this configuration is often known as the modern or digital colour palette, sometimes even the RGBCYM[K] palette.

You can then fill in the intermediate segments with new secondary colours to get a simple 12 colour wheel again. Amazingly if you employ the idea that opposite colours can be complimentary or adjacent colours analogous and you find they are also harmonious. What is going on here is the traditional colour wheel wrong? Perhaps for both their photography and printing aspects digital photographers might be wise to adopt or at least consider this new colour wheel. I will be discussing many of the issues for digital photographers in coming blog posts.

The very last image shows the colours Adobe Photoshop and Lightroom give you control sliders over. I’m still no closer to being able to explain this colour selection.