In this paper, a new technique is presented for unsupervised change detection in color satellite multi-temporal images. It is accomplished by three stages. First, the color image is modeled into the pure quaternion image. In order to process color image directly, the quaternion mode is introduced, in which the Red, Green and Blue values of each pixel of a color image are represented as single pure quaternion valued pixel. And a color image can be represented as a pure quaternion image. Second, a computationally simple method for singular value decomposition (SVD) is adopted to perform principle component analysis (PCA) on the pure quaternion image. Finally, change detection in a pair of color satellite images can be achieved. Experimental results show the superior of the proposed method over the state-of-the-art methods for change detection.