In the course of vehicle license plate (VLP) automatic recognition, tilt correction is a very crucial process. According to Karhunen-Loeve (K-L) transformation, the coordinates of characters in the image are arranged into a two-dimensional covariance matrix, on the basis of which the centered process is carried out. Then, the eigenvector and the rotation angle α are computed in turn. The whole image is rotated by -α. Thus, image horizontal tilt correction is performed. In the vertical tilt correction process, three correction methods, which are K-L transformation method, the line fitting method based on K -means clustering (LFMBKC), and the line fitting based on least squares (LFMBLS), are put forward to compute the vertical tilt angle θ. After shear transformation (ST) is imposed on the rotated image, the final correction image is obtained. The experimental results verify that this proposed method can be easily implemented, and can quickly and accurately get the tilt angle. It provides a new effective way for the VLP image tilt correction as well. According to Karhunen-Loeve (KL) transformation, the coordinates of characters in the image are arranged into a two-dimensional covariance matrix, on the based on which the centered process is carried out. Then, the eigenvector and the rotation angle α are computed in turn. The whole image is rotated by -α. In the vertical tilt correction process, three correction methods, which are KL transformation method, the line fitting method based on K -means clustering (LFMBKC), and the line fitting based on least squares (LFMBLS), are put forward to compute the vertical tilt angle θ. ST) is imposed on the rotated image, the final corrected image is obtained. The experimental results that that proposed method can be readily implemented, and can quickly and accurately get the tilt angle. It prov ides a new effective way for the VLP image tilt correction as well.