Shadow detection for high spatial resolution remote sensing images is very critical for image segmentation

feature extraction

image matching

automatic target detection and target location. In order to improve the accuracy of shadow detection

we propose a new shadow detection method based on a statistical mixture model

which combines several radial basis function neural networks. Four statistical features

including energy

entropy

contrast and inverse difference moment

extracted from grey level concurrence matrix are used as the model input features. EM-like algorithm is adopted to estimate the model parameters through optimizing the system cost function. Comparative experiments are performed between the Gaussian background model and the histogram threshold method. Experimental results show that higher detection accuracy of the proposed approach is obtained. The proposed method can solve the problem such as high reflective regions and false alarms in the presence of water

as well as the repeated threshold calculation.