Object-Based Correspondence Analysis for Improved Accuracy in Remote Sensing Change Detection

The correspondence analysis(CA) method,a multivariate technique widely used in ecology,is relatively new in remote sensing.In the CA differencing method,bi-temporal images are transformed into component space,and individual component image differencing can be performed to detect potential changes,somehow similar to principal component analysis. The advantage of the CA method is that more variance of the original data can be captured in the first component than in the PCA method.However,these techniques are all performed on a pixel by pixel basis,becoming unsatisfactory in some circumstances due to higher spectral heterogeneity in imagery of high spatial resolution.This problem can be alleviated by the object-based strategy,which segments the image into regions of relative homogeneity,which are,in turn,used as the basic units for data analysis.We proposed an object-based CA approach for change detection,whose performance was compared with those of pixel-based PCA and CA.Experimental results show that the object-based CA method produces the best accuracy in change detection.