Heatmap / Continuous Colour Scale Gradient Palettes

After previously implementing support for image value remapping in my image-processing infrastructure, I’ve been looking into generating better gradient colour mapping schemes / palettes than the initial hard-coded HSV ramp I used.

The thermal camera I have can bake in one or two pretty nice presets, but as previously discussed, I wanted the freedom and flexibility to “tone-map” them myself afterwards, for more creative freedom.

There are a wide range of colour schemes in various plotting and visualisation libraries and tools, with different schemes being either more- or less-suitable depending on the use-case and the values being represented. In particular, having a lot of different strong colours in a gradient can make the resultant mappings look pretty or striking, but it is then often less easy to perceive the magnitude of values compared to more simpler gradients – but on the other hand does have the advantage that any value differences are mapped to a more obvious change in colours.

More recent/modern colour mapping schemes have aimed to improve aspects such as being perceptually uniform, and being more friendly to those suffering from colourblindness, whilst still producing visually appealing results.

The docs page for Plots in Julia has a nice set of example colour schemes / palettes, so I decided to try copying some of the more interesting colour mappings from those schemes to see how they looked, essentially by just interpolating splines of linear colour values to evaluate the gradients, which for my purposes should be sufficient.

Example gradient strips:

Hue (reversed):

HueSV gradient ramp

At first glance, this gradient palette that varies the Hue colour looks nice, but the similarities in redness at either end (red and magenta), due to the Hue of HSV being a circular map (or mapped to a cone or cylinder) mean that the colours at the extremities are somewhat difficult to tell apart, especially for those who are affected by colour blindness. It’s also more difficult to interpret a “rainbow” of colours to gauge their magnitude, so in practice this gradient doesn’t seem that suited for many visualisations.

Heat:

Heat gradient ramp

This is very effective and simple for both thermal images and density plots / maps, and is essentially a limited, rough approximation of “black-body” temperature emission colour (it stops at white, rather than continuing on to blues).

Inferno:

Inferno gradient ramp

This is also very effective, and seems to closely match one of the presets my thermal camera can bake into its images.

Jet:

Jet gradient ramp

A colourful gradient that has better colour separation at either end than Hue, although still likely suffers from the “too many colours” issue for the use-case of gauging value magnitude compared to more simpler schemes.

Plasma:

Heat gradient ramp

A more limited gradient.

Turbo:

Heat gradient ramp

Visually quite pleasing, but again, as this was largely designed for visualising disparity and depth maps, it isn’t primarily aimed at being able to easily gauge value magnitude, but is an improvement over Jet.

Viridis:

Heat gradient ramp

A very pleasing colour gradient that seem to work well for density plots.

Thermal Image examples:

Collage of example thermal image colour schemes Clockwise from top-left: Hue, Heat, Inferno, Turbo, Plasma, Jet.

Shipping density map plot examples:

Collage of example shipping density map plot colour scenes Clockwise from top-left: Heat, Inferno, Jet, Viridis, Turbo, Plasma.

So far, from the colour gradient schemes I’ve played around with, I think Heat and Inferno are my favourite schemes for thermal images, and they seem to additionally work very well for density plots and maps too. Viridis may also become one of my go-to colour gradient schemes for density plots, as that appears to also produce nice results.



Image Value Remapping

I’ve finally got around to finishing pretty comprehensive support for “remapping” pixel values from images within my image-processing infrastructure, so as to be able to scale or remap pixel values in an image to another representation.

It supports both “linear scalar” scaling, as well as scalar-to-colour (i.e. to an HSV gradient ramp) and scalar to custom configurable colour gradients, in addition to being able to do additional pre-remapping scale and fit value modifications - i.e. re-fitting values from a data value input range of [-5000.0, 20000.0] to [0.0, 1.0] first and then applying scalar-to-colour gradient remapping afterwards.

This was mainly for two particular use-cases I had: the first was some more map visualisations I’m playing with - in particular some data on Global Shipping Traffic Density from the World Bank, where I got a bit frustrated with QGIS’s gradient-remapping functionality, so decided to roll my own in order to allow it to be more recipe-based with configurable parameters and then allow remapping multiple images with exactly the same params more easily.

Below is an example of remapped Global Shipping Traffic Density around Europe, with the reprojected (in terms of map projection) absolute values from the source file data being remapped to a custom colour gradient output image:

Remapped image data (from World Bank Group) of shipping traffic

The second use-case was remapping 8-bit greyscale values in JPEG files taken from a thermal camera I’ve had for a while, to be able to remap them to more interesting and configurable colour gradients.

Original thermal camera image:

Thermal camera image

Remapped image:

Remapped termal image

Original thermal camera image:

Thermal camera image

Remapped image:

Remapped termal image

The thermal camera I have does support some colour gradient automatic remapping itself, however there are only 8 presets, and the result is baked into the captured JPEG image, so having the flexibility to capture in greyscale in the camera and then remap the images later is much more useful.

I do have some older interesting images I captured just after getting the camera years ago, where I was mostly capturing the images in pre-baked gradient format, which I would like to change the gradient of - so a possible feature improvement to my image-processing infrastructure might be to add support for “reverse-mapping” the colours in images from a gradient back to a “scalar” or linear value, so I could then apply custom colour gradients of my choosing.




Archive
Full Index

2024 (6)
2023 (7)
2022 (3)
2021 (5)
2020 (4)
2019 (7)
2017 (1)
2016 (2)
2015 (1)
2014 (9)
2013 (10)
2012 (7)


Tags List