Adaptive Watershed Segmentation of Remote Sensing Image Based on Wavelet Transform and Fractal Dimension

considering the over-segmentation problem of traditional watershed segmentation, we proposed an adaptive watershed segmentation method. We made four improvements in this paper. Firstly, by using multi-level wavelet transformation and selecting different thresholds and filter functions corresponding to different frequency components, random noises could be eliminated effectively in the image. Secondly, we proposed a new method of markers extraction based on scale, gradient and edge information. The true markers which are relevant to objects can be extracted precisely from the homogeneity gradient image. Thirdly, we proposed the cost of region expansion was calculated based on spectrum features and fractal dimension. The regions can expand with inner homogeneity as far as possible. Fourthly, we designed a new algorithm of adaptive threshold selection, and proposed a heuristic decision of region expanding based on it. Our heuristic decision can make regions expand in a simultaneous way. From the experiments, the improved adaptive segmentation method can make regions expand not only in a simultaneous way but also with inner homogeneity as far as possible. The proposed method can obtain meaningful and homogeneous regions with accurate, consecutive and one-pixel wide boundary.