Scale and Rotation Invariant Recognition Method Using Higher-Order Local Autocorrelation Features of Log-Polar Image

This paper proposes a scale and rotation invariant recognition method which uses higher-order local autocorrelation (HLAC) features of log-polar image. Linear scalings and rotations are represented as shifts in the log-polar image which is obtained by re-sampling of the input image. HLAC features of log-polar image become robust to the linear scalings and rotations of a target because HLAC features are shiftinvariant. By combining these features with a simple classi er which uses linear discriminant analysis, we can design a scale and rotation invariant recognition system. Robustness to the scalings and rotations are conrmed by experiments on 2D shapes and face recognition. Robustness to the changes of backgrounds is also con rmed by experiments on face recognition.