Image Based Smoke Detection Using Source Separation

In this paper, we propose an automatic image based smoke detection using source separation. In particular, we assume that the region of interest (smoke region) is a linear combination of smoke and background pixels, and we estimate the smoke component. More specifically, we extend the linear hyperspectral unmixing techniques to the context of image based smoke detection in order to separate the smoke component from the background. The proposed approach yields promising results especially with smoke images captured outdoor.