Abstract
Damage due to fire can be minimized by discovering fires early. Fires produce smoke as well as flames, so detecting smoke from data like surveillance camera images data can make it easier to discover fires early. However, in attempting to discover fires at locations which are hard to see with the human eye, as in cases like forest fires, we are faced with a problem in that smoke does not have a determinate form in camera images. It has a pattern similar to features like clouds, sky and forests, and thus is difficult to detect with processing approaches like simple background subtraction or pattern matching. So this paper proposes a technique for extracting smoke regions from an image using fractal encoding concepts. We were led to this idea by noticing that smoke shapes have the property of self-similarity, and here we attempt to extract smoke regions by discovering the distinguishing features of smoke regions in the code produced by fractal encoding of an image.
