Segmentation for Eyed Typhoon Cloud Image by Curvature and Fractal Feature

Abstract

An efficient method to segment eyed typhoon from a satellite cloud image is proposed. First, original satellite cloud image is enhanced by gray transform. Second, in order to reduce the computation burden to segment the whole satellite cloud image, single threshold segmentation based on Bezier histogram is implemented to the original satellite cloud image. Some small unrelated cloud masses are discarded. Third, a second segmentation based on Bezier histogram is carried out to the first-segmented typhoon cloud image in order to completely separate true typhoon region into other non-typhoon regions. Box dimensions of all the regions in the second-segmented cloud image are computed. The region whose box dimension firstly arrives to the valley in the box dimension curve, which is drawn by all the box dimensions of second-segmented cloud image, is identified as true typhoon region. Forth, the second-segmented typhoon cloud image is expanded in order not to lose some important details. The second-segmented threshold is combined with Bezier histogram of expanded cloud image to determine the optimal third-segmented threshold. The third segmented cloud image should include most of important information of typhoon. Fifth, discrete stationary transform is implemented to curvature curve of Bezier histogram of the third-segmented cloud image. Multithreshold segmentation is implemented to the third segmented cloud image in stationary wavelet domain. Final segmented typhoon cloud image is obtained by selecting an optimal segmentation scale in stationary wavelet domain. Experimental results show that the proposed method can efficiently segment the typhoon cloud series from the satellite cloud image. The new method is better than Olivo method and H.Q. method.