Color sampling is the process of taking color information at certain positions (sample points) on an image to use it for various applications. For instance, the brightness can be a driver for a perforation or displacement pattern on a facade. It is also the basis of texture mapping for rendering. This process can also be used directly to bring an image over an architectural element like a panelized facade by unfolding the facade geometry onto the image and sampling colors of each panel. However, depending on the material, fabrication process, design intent, budget, etc. this process might require some additional steps to meet the required constraints. The following is a process we designed to map an image onto a panelized wall so each panel has only one color, the total number of colors is limited to a specific number, and the colors are based on a standard pallet e.g. Pantone.
Step 1. Alignment and Sampling
The process starts with overlaying the panels onto the image. We normalize (scale to 0 to 1) the boundaries of both the image and the facade to guarantee that all panels will be within the boundary of the image. Then we specify 9 points for each panel and sample the color for each point. We take the average of the nine colors and assign it to the panel. Sampling multiple points for each panel helps bridging the color shift from one side of the panel to the other.
Step 2. Clustering Colors and Reducing Color Count
The result of the previous step are panels each with a unique color. In this case we want to limit the number of unique colors to a specific count. Using K-Means machine learning algorithm we cluster colors with their RGB values and for each cluster we use the average color of the cluster for all the panels inside it. This can significantly reduce the color count.
Step 3. Convert to Standard Pallet
In various scenarios fabrication can benefit from using a standard color pallet instead of random RGBs. In this step we search though a standard pallet e.g. Pantone, to find the closest color to each color from the result of the previous step.
Different color attributes can be used to find the closest color, and each has its own pros and cons.
We used weighted average of normalized Hue, Saturation and Value attributes since it gives us the most control over what we prioritize to find the best match.
Step 4. Final Adjustments
Although we have maintained the overall look of the image through the rationalization process, naturally the image will lose some of its qualities. In this step we include a couple of methods in which we can bring back what we might see too important to lose. These steps intentionally involve some manual work from the designer to allow creativity and artistic vision create the final outcome.
Step 4.1. Color Replacement
Changing a color from previous step to another color from the standard pallet. In this example changing Illusion-Blue to Blanc-De-Blanc from Pantone pallet brings back some of the highlights.
Step 4.2 Adding Color
The clustering and reducing color count will average out the colors and lowers the contrast. To bring back some of the sharp colors we can chose the color we want, specify the area and a ratio and the tool randomly chooses a percentage of panels in that area and assigns the specified color to those panels.
