I’ve been thinking about how colour could be displayed for a Satellite Image. The Earth is beautiful. In some software, in the legend, a Satellite image is displayed as below
Image from http://webhelp.esri.com/arcgisexplorer/2500/en/legend_window.htm
It does contain some useful information, ie what band is being displayed Red, Green and Blue, but it tells us very little about the image. In the past I have often skipped adding this to a legend; I am not sure it adds much value to the viewer of the map.
Legends like the one above (taken from here) are very informative but it would be nice to get a visual distribution of quantities, or at least an approximation. Perhaps something like the images below.
These two colour bars are from adapted from a Sentinel 2 image on Lipari: the first is a true colour RGB satellite image and the second is a false colour RGB image.
This image is based on Sentinel 3 data over a clear day in over France in March 2017. 5 different greens… who knew?
These colour bars could be added to maps, potentially providing a more meaningful image. Does it cause you to look more closely at the data? Are you surprised by the distribution of the colours?
The dominant colours shown above are perhaps not quite as expected (a brighter pink seems missing in the second image for example). I have used k-means clustering to separate a user defined number clusters from the image. Pixels in one cluster will be more analogous (wrt color) than pixels in another separate cluster. There will be a slightly different result if I ran the code again, this is an unsupervised classification process after all.
I think it produces a pleasing visual, that the eye is drawn to and gives the viewer a distribution of the colours. I am not saying these are great cartographic products, just an interesting way of presenting the beautiful colours that satellites can give us over the range of spectral bands.
A bit of open cv, a bit of sckit learn and a bit of PIL…
I was inspired to write this blog post after reading a great blog on pyimagesearch based on dominant colours. I have slightly adapted the code to scale and order the colours and crop the image in PIL, but if you follow the tutorial you to can create your own sat colours!
I can’t see a reason why not to try this on any type of raster display.
Maybe you will find a colour that inspires you, will your new kitchen be a Wetland Blue, or the bathroom a Saharan Orange? 🙂