A method, referred to as Composite of Central and Ring Projection (CCRP), is proposed to extract features with rotation invariant property. It reduces the dimensionality of a two-dimensional pattern by performing both central projection (CP) and ring projection (RP). A dissimilarity function is developed and used to distinguish different patterns. This function makes use of both similarity corrections of RP and CP. Information along both circles and polar angles can be retained from the original pattern. Some experiments have been conducted, in which a set of ambiguous printed Chinese characters which are partially damaged or polluted by noise were classified. The experiments have satisfying results. © 2009 IEEE.