Aiming at a large amounts of computation existing in attitude measurements, a new method of 2-D object attitude measurement based on a covariance matrix is presented. Firstly,an image is segmented by the threshold and the center of an object is extracted. Then,the covariance matrix of the object is computed and decomposed by a orthogonal decomposition,and the elliptic equation defined by the covariance matrix is gotten. Finally,the lengths and orientations of both axes are calculated to obtain the scale change and rotation between the two images. The attitudes of objects in 256 pixel×256 pixel images are calculated with a log-polar algorithm and the presented algorithm respectively. The experimental results show that the angle error is within ±1°, and the scale error is within ±2% with proposed method,which is superior to those computed by the log-polar method, while the computation time is only 1/16 that of the log-polar method. The presented method can calculate the rotation and scale change fast and accurately, and can meet the requirement of real-time processing.