Estimating Object Rotation And Scale Using Correlation Filters

Recently, much attention has been paid to designing correlation filters that provide distortion-invariant pattern recognition. However, optical correlators can also be used to estimate the orientation and scale change in the object being observed. This paper presents two new methods for designing correlation filters that are aimed at determining the in-plane rotation and scale changes present in the input image (relative to a reference image). These methods assume that a known object is present in the input plane and that its rotation and scale are to be estimated. The first method is capable of determining both rotation and scale changes simultaneously. The second method assumes that the scale is fixed and estimates only the in-plane rotation. Both methods should prove useful in applications in which one needs to estimate the object orientation. Several simulation results are included to quantify the estimation accuracies and the computational complexities of the two methods.