A new rotation estimation and recovery algorithm

In this paper, we present a new rotation estimation and recovery algorithm. The algorithm recovers the rotated images with high quality through minimizing the number of pixels whose intensities need to be altered. The recovery algorithm can tolerate up to about +/-4 degrees error in the estimation of the angle of rotation. To estimate the angle of rotation, a non-rotated version of the rotated image is used as a reference. Three different methods to estimate the rotation angle are discussed; image alignment, radial Tchebichef moments, and Fourier descriptors. Image alignment shows the most accurate results for the estimated angle. The results of our recovery algorithm are compared more favorably to those of another algorithm that uses interpolation (Matlab built in rotation function).