Many computer-assisted classification algorithms have been developed to produce the land-cover maps from multi-spectral satellite images. Since the influence of the atmospheric conditions and the change of the earth surface, most of the image classification methods are image-dependent. For instance, a great number of supervised classification algorithms normally can not apply the same training data to multi-temporal image classification. As a result, the training data has to be re-selected image by image in order to perform multi-temporal image classification. The purpose of this study is to develop an image-invariant classification technique to classify multi-temporal satellite images.