Visualizing Color Usage

TLDR:

Our goal of this post is to implement color quantization in R, compare several methods of detecting segments of an image, and start creating the visualization UI. It’s no surprise that this topic naturally lends itself to some preliminary data exploration and visualization, and R makes generating some really cool graphics easy.

Implementing Color Quantization in R

Quantization involves expressing a range of values by one quantum value. In images, it is the
quantization of color spaces, such as RGB, CMYK,HSV, etc. RGB is one of the most popular color representation models. Each pixel has three channel values: red, green, and blue. Each channel has a value for its intensity (e.g., an integer between 0 and 255 for an 8-bit color representation). An image’s pixels are rendered by combining intensity values from three RGB channels to produce a color value.

In color quantization, this is achieved by reducing the number of colors and compressing an entire color range into one specific color. Using Edvard Munch’s famous Scream as an example, we applied k-means, median-cut, and hierarchical tree-based algorithms to partition an image and extract the top n dominant colors using the following:

• kmeans – We built the k-means clustering algorithm using the kmeans function in the stats package.
• median-cut – We used image_palette function in the RImagePalette, which is a pure R implementation of the median cut algorithm.
• Oct-tree based – We used image_quantize function in the magick package. Based on its oct-tree algorithm, we built a function get_my_colors to get k number of colors of the image and returns the top hex values for simpler visualization.

K-means Algorithm

We applied k-means to our data frame using different k numbers. Replacing the color of each pixel in the image with the mean R, G, and B values of the cluster in which the pixel resides yields the following results. Figure 3 below shows the segmented image when k=2 and k=6, i.e., breaking the image into 2 color regions or 6 color regions.