A pattern matching method using geometric information of images

We present a hybrid method for image pattern matching using gradient and geometric information of images. The gradient orientation and the vector length allow us to estimate the rotation angle parameter and the scale parameter over two 1D search spaces, respectively, thus reducing the computational complexity. The correlation based method can be applied to obtain the translation parameters accurately when the angle and the scale parameters are already known. The decomposition of a 4D search space gives us a practical way to deal with the similarity transformation between images. Our experiments show that the combination of geometric feature and intensity information in the images significantly reduces the computational load and makes the matching accurate and robust compared to the conventional intensity-based methods.