Research on Auto Extraction of Interested Region in Fire Image Compression

Forest fire detection includes satellite monitoring, manual monitoring, aerial survey and so on. Compared with these expensive ways, a more practical fireproofing surveillance system needs to be built. An image capture system can be set up in the tower to sent fire image back to the command centre, using the short wave radios equipped for manual monitoring in the towers. Considering the finitude of short wave bandwidth, it is essential to compress the image. Region of interest coding can provide higher compression rate. In this paper, the image zoning is implemented using watershed and Normalized Cuts. Then, the mask of region of interest is extracted aiming at the features of fire. In order to improve the speed of calculating, zoning is done in the subband of image.