An approach for extracting smoke plumes using entropic thresholding and region growing is proposed in this paper. First, the stack is located by an edge detector in an image. The stack location is taken to be a basic position for finding the plume. Several threshold values are then calculated using entropy measurement of the image based on a co-occurrence matrix. These threshold values are used to threshold the image into a lower-level image. A region-growing technique is applied to segment the thresholded image. Further processes are used to refine the segmented result. Finally, several features of the regions are derived using principal component analysis to extract the plume in the segmented image. We apply the approach to several examples to examine how it performs. The results indicate that the approach is a promising technique for plume extraction.
