“…In conclusion, texture is useful, but there are two difficulties with the approach: (i) sky has no texture, so that the borders of cloud extend into regions of sky causing a small overestimation of cloud; (ii) some regions of cloud have little texture and are indistinguishable from sky, they are therefore consistently mislabelled (see regions of cumulus in left and middle image of Figure 5). 4 Using colour and texture Both colour and texture have been shown to be important indicators to the presence of cloud, but neither is sufficient for complete segmentation. Chromatically, haze is indistinguishable from cloud, and cirrus is indistinguishable from sky.…”