Abstract:

In order to make full use of spectral and texture features, an object-oriented change detection method for remote sensing images based on multi-features fusion is proposed in this paper. First image segmentation is used to get image objects. Then the spectral and lbp texture histograms of each object are extracted. G statistic is adopted to calculate the distance of histograms between two periods. The heterogeneity of each object is built by weighted spectral and texture distance. At last, the expectation maximization algorithm and bayesian rule with minimum error rate are applied to get the change/no change results. Experimental results on QuickBird and SPOT-5 images show that the method proposed in this article can integrate the spectral and texture features effectively and improves the accuracy of change detection.

Key words: object-oriented, multi-features fusion, local binary patterns, G statistic, expectation maximization