An Adaptive Threshold Segmentation Method Based on Spatial Statistic Theory to High-resolution Remote Sensing Change Detection

An adaptive threshold segmentation approach is proposed for multitemproal high-resolution remote sensing change detection. The threshold segmentation approach is based on the spatial statistics theory,using kriging interpolation. Taking differencte images as input data,it is classified as changed,unchanged and uncertainty classes. In view of uncertain objective pixel,we use changed and unchanged pixels in local area as samples to estimate the objective pixel. Then we compare the actual value and target value of objective pixel to determine the types of uncertainty pixels.To test the advantages and adaptability of this method,a series of traditional threshold segmentation algorithms have been carried on comparctive analysis and multitemporal high-resolution remote sensing images are introduced as experimental data. Experimental results prove that the adaptive threshold segmentation in this paper can remove false changed information,reduce false alarm and improve detection accuracy by comparing with traditional threshold segmentation algorithms.