HIGHLIGHTING A CLOSE-TO-ONE-SPECIFIC-COLOR SURFACE BASED ON THRESHOLDING COLOR-CHANNEL-WISE DISTANCES TO A PIXEL PALETTE

Abstract

A conception of highlighting close-to-one-specific-color surfaces (sea water, a green forest, asphalt
on a highway, etc.) is presented. The goal is to develop an efficient algorithmic routine for highlighting such
surfaces. Once a bank of diverse images having the color of interest is chosen, a pixel palette is formed. The pixel
palette is of one-pixel samples, each of which has a unique color close to the color of interest. Then, for every
single pixel sample in the palette, color-channel-wise distances between the image and palette are computed and
normalized. The normalized distances are compared to a threshold for every sample. If the distances are less than
the threshold (for all the color channels simultaneously), the corresponding pixels of the image are highlighted.
With the before-formed pixel palettes, when the threshold is fine-tuned, the developed routine is fully automatic.
It requires only a type of highlighting task (whether it is the sky, clouds, asphalt, calm river water, or any other
similar surfaces of medium complexity) to be input to load the corresponding pixel palette and threshold. Similar
to chroma keying, the suggested method can nonetheless work with complex colors, so it is more robust.