A transformation invariant matching algorithm for handwritten chinese character recognition

Abstract

All the Chinese characters are composed of some fundamental characters called radicals. The algorithms we present here are used to match the radicals, stored in the data base, with the input Chinese character. From the radicals in the input Chinese character, we can know what this input Chinese character really is. The matching process is equivalent to checking the validity of the relation that the radical is a sub-character of the input Chinese character. The least-square-error estimation and some properties of the Chinese characters are also used here to check whether the relation mentioned above is satisfied, and if this relation is satisfied, the sub-character matched with the radical is extracted. This matching algorithm has some superior merits which are (1) transformation (scaling; rotation and translation) invariant and furthermore (2) invulnerable to the inherent defects of the thinning algorithms. It overcomes these defects at the expense of more computation time. But it can be run in parallel, so large computation time is no longer a very serious problem. Experiments with 114 characters and 62 radicals have been conducted, and we find that the success rate of our matching algorithms is above 99%, thus we conclude that our algorithms are very reliable.