An operational method to determine change threshold using change vector analysis

摘要

Digital change detection (CD) is the computerized process of identifying changes in the state of an object, or other earthsurface features, between different dates. During the last years, a large number of change detection methods have been proposed for change detection of multiple-temporal remote sensing images. Among these, change vector analysis (CVA) is a very important and widely used method. The key of CVA is to determine change detection threshold. Change detection threshold is a very valuable key for change detection precision. In the literature, many techniques to determine change detection threshold have been proposed. However, most of them are not robust and operational since images are diverse and complex, especially to very high resolution (VHR) data (e.g. images acquired by QuickBird, IKONOS, SPOTS and WorldView satellites). Such discrimination is usually performed by using empirical strategies or manual trialand-error procedures, which affect both the accuracy and the reliability of the change-detection process. In this paper, we analyze the algorithm based on minimal classifying error, the algorithm based on OTSU and the algorithm based on EM. To eliminate the complexity of VHR data, an improved algorithm based on EM is proposed. Suppose the difference image meets the Mixed Gaussian distribution model. First, the grey histogram of the difference image is fitted to the Mixed Gaussian Distribution Model (MGM). Then the change detection threshold is determined by the MGM graph combing the Bayesian Criterion and the actual situation. In experiment, the semi-automatic method is effective and operational.