Efficient Dimensionality Reduction using Principal Component Analysis for Image Change Detection

Change detection in image processing is the process of identifying differences by comparing images taken at different times. There are several digital change detection techniques; nevertheless, there is no universally optimal change detection methodology: the choice is dependent upon the application. Change detection methods based on multispectral space transformations like Principal Component Analysis (PCA) show good solutions for remote sensing applications. One advantage of PCA is in reducing data redundancy between bands and emphasizing different information in derived components. This work focus on the PCA exploitation for the SPOT multispectral image change detection. Thresholds are applied to the transformed image (PC2) to isolate the pixels that have changed. Thresholding methods require a decision as to where to place threshold boundaries in order to separate areas of change from those of no change. The accuracy of change detection maps that are derived with SPOT data is represented in terms of producer’s accuracy, user’s accuracy, and overall accuracy, which are calculated from an error matrix (or confusion matrix). The obtained results have demonstrated solving efficiently the change detection problem.