Watershed Segmentation Algorithm Based on Gradient Recovery and Iterative Minima Calibration

Segmentation is the primary step and an important link in object-oriented remote sensing image classification. However, traditional watershed segmentation algorithms often lead to over-segmentation, causing the segmented ground feature image objects to be overly fragmented. It is found that the watershed segmentation algorithm can be improved from two aspects: the calculation of gradient by using differential operators and region merging. When performing gradient operations by using the Sobel operator, the entropy values of each band are introduced, and the ratio of entropy value of each band to the total entropy value of all bands is used as the gradient weight for gradient recovery. Meanwhile, for the improvement of the minima calibration for region merging, after calibrating the foreground and background, iterative calculation of the foreground mean and background mean is substituted to obtain the minima threshold and complete the image segmentation. Through comparative analysis of the results, this method not only solves the problem of over-segmentation well, but also ensures the edge information of ground features to the greatest extent possible, providing better support for subsequent analysis and detection of ground feature objects.