Hue alone is often not enough for reliable color detection. In dark or weakly saturated areas, hue becomes unstable or effectively meaningless, so pixels in shadows or near-neutral tones can produce misleading results.

From the shared answer, the source image’s hue channel itself does not appear to contain the banding or the false matches you described, which suggests the issue is more likely in the processing/math used after extracting hue than in the camera hardware or compression.

Also remember hue is circular: red is near both 0 and 255/360, so hue distance must be computed as an angular wraparound, not simple linear subtraction. If your comparison doesn’t account for that, you can get incorrect “closeness” values.

For better color isolation, combine hue with saturation and brightness/lightness thresholds. A common approach is: only trust hue where saturation is above a minimum level, and optionally exclude very dark pixels. That usually removes many false positives.

So: you probably don’t misunderstand hue fundamentally, but hue by itself is not sufficient, and your implementation likely needs wraparound-aware hue distance plus saturation/value filtering.