The linear algebraic analysis of remotely sensed multispectral image is studied. The inner product is applied to the classification and the outer product to systematic treatment of mixels. The analysis uses essentially angular distribution of data with respect to the origin of feature space. The offsets which are involved in the data are estimated by the 3-dimensional histogram and subtracted from the data. Then the fact that pixels have the same angle in the feature space strongly implies that they have the same reflectances and belong to the same category. Therefore the angles of data is the most important information to classify the categories.
