Segmentation for Satellite Cloud Image by Combining Fractal Features and Continuous Wavelet Transform

Abstract

A kind of algorithm to segment typhoon cloud image is proposed. First, fractal feature of typhoon cloud image is used to recognize center dense cloud region so that an initial segmentation region (ISR) including typhoon cloud series can be determined. Morphological operator and median filter are used to reduce latitude lines and amplitude lines in the ISR so as to obtain the pre-processed ISR (PISR). In order to reduce noise in the satellite cloud image, Bezier curve is used to smooth histogram of ISR. An optimal initial segmentation threshold is determined from multiple thresholds obtained from Bezier histogram. The optimal threshold is used to segment the PISR. The maximum region in the segmented PISR is selected as region of interest (ROI). Other small cloud masses are reduced by the method. We replace the ROI with the original typhoon cloud image to obtain the segmenting ROI (SROI). Again the Bezier histogram is used to smooth the noise in the histogram of SROI. In order to accurately detect the peaks and valleys in the curvature curve of the Bezier histogram, continuous wavelet transform is used to accurately recognize the location of peaks and valleys. We will obtain multi-segmented images at different scales. A criterion is employed to select an optimal segmentation scale. Finally, the whole typhoon cloud series is accurately segmented by the method. Experimental results show that the proposed algorithm can efficiently segment the typhoon cloud series from the satellite cloud image. It is better than Olivo method andH.Q. method.