This paper presents a property-based method defined on principal components for automatic shadow detection, called the PCA method. This method applies both RGB and HSI color models. In addition, some shadowed image patches still offer the brightness and color information for shadow compensation. This research compares the methods of histogram matching and local statistics method, and figures out the effectiveness of the two methods. Experimental results are evaluated in terms of subjective and objective evaluation figures. Test results demonstrate that the proposed PCA method with the input data of the four bands R, G, B, and NIR provides the best accuracy of shadow detection, e.g. with the overall accuracy of 93.7% and 92.2% in the test areas A and B, respectively.