Image Sampler with rgba images

Replies to This Discussion

Permalink Reply by David Rutten on May 26, 2017 at 8:49am

There's two major approaches you can take, and probably a whole bunch of alternatives I haven't thought of:

1. Create a gradient that closely matches the image gradient, evaluate this gradient at a whole bunch of positions, then use the [Find Similar Member] component to figure out to which gradient colour a sampled colour is closest. You then need to convert the index of the most similar colour back somehow onto a sphere radius.
2. Pick a few well-spaced colours that represent important 'keys' in the gradient. Convert them to points (rgb -> xyz). For each sampled colour, figure out the distances to each key colour. Map this distance onto some inverse-equation (1/(x^2) for example, or use a Graph Mapper), and use these mapped distances as weights for a weighted average computation.

Neither of these approaches is particularly straightforward, but I recommend starting with the first as it a more direct algorithm (you always work with colours, making it more understandable).

Permalink Reply by Joseph Oster on May 26, 2017 at 11:40am

The problem, as I see it, is mapping the colors to the sequence you wish, with red returning the largest value and blue the smallest.  I tried several different approaches, including Color Hue instead of RGBA, dividing the domain into five ranges (colors) to 'ReMap', etc.  Finally ended up with an expression to evaluate RGB values, but none of the methods really worked well.  Used circles instead of spheres for better performance while testing.

Attachments:

• img_sampler_2017May26a.gh, 784 KB

Permalink Reply by Joseph Oster on May 26, 2017 at 4:10pm

I tried following David's suggestion #1, with only limited success.  Still not happy with the mapping to correlate color to diameter...  Instead of comparing RGB colors with 'FSim (Find Similar Member)', I had better luck converting RGB values to a single integer value (cluster 'ColorN' below), then using 'FSim':

But there are still anomalies.  :(  The 'colors' slider divides the gradient.

Attachments:

• img_sampler_2017May26f.gh, 798 KB

Permalink Reply by Joseph Oster on May 26, 2017 at 7:49pm

Here is an effort at David's suggestion #2, using RGB values as points (cluster 'ColorP') and distance.  No "inverse-equation" involved.  Instead of a gradient, I used five 'Color Swatch' components set to color values plucked from the original image using GIMP.  They are used primarily to determine which diameter, from a list, is used for each color.  The distance between each sample point and all five 'ColorP' points is computed, sorted, and the smallest values are used (color and associated diameter).

The blue group has a switch to choose "Sample" colors for all the circles, matching the image, or "Map" colors (one of the five plucked colors):

Sample:

Map:

Attachments:

• img_sampler_2017May26g.gh, 798 KB

Permalink Reply by Joseph Oster on May 26, 2017 at 9:39pm

'Circle' and 'Boundary' replaced with 'Sphere' at 0.05 resolution (1/20th of actual image):

Permalink Reply by Paolo Tamburini on June 12, 2017 at 9:37am

Hi, sorry for the late reply.

Wow it is perfectly what I was asking for!! so thank you!

But I didn't understand for what you use the Sort component. I think I don't understand what this component does. 

Permalink Reply by Joseph Oster on June 12, 2017 at 9:53am

17 days... ancient history, my friend!

Permalink Reply by David Rutten on June 12, 2017 at 11:32am

Sort operates on a list of numbers (nothing else, just numbers) and it will put them in a smallest->largest order. It has additional functionality though, in that you can use the number sorting to synchronously sort any other list of data.

For example imagine you have a collection of curves, and you want them ordered shortest to longest. Step one is to measure the length of each curve. You now have a list of numbers which is equally long as your list of curves, and with corresponding value ordering. These measurements are then sorted using the K input of the sort component while you plug your original curves into the A input. As the numbers are shuffled to go from lowest to highest, your curves will simultaneously also be shuffled in the same way.

Permalink Reply by Joseph Oster on June 12, 2017 at 1:22pm

Thanks David.  I had to re-open the code to remember how I was using it here.  In this case, I'm sorting the "distance" between the sampled RGB color (converted to a "point") and each of the five primary colors in my list.  That sort by distance is used to sort the five colors and their associated diameters, then the first item in each list (corresponding to the shortest distance) is used for that "pixel" in the image.

I can't quite fathom the 17 day delay in looking for replies, though?

Permalink Reply by Paolo Tamburini on June 13, 2017 at 4:42am

ok great, I understand. Thank you David, thank you Joseph!