Broken kannada character recognition — A neural network based approach

Character degradation is a major problem in character recognition. Most of the historical documents are degraded due to extrinsic factors like blurring, skew, background noise etc and intrinsic factors like distortion, broken characters, touching characters etc. In this paper we propose a novel approach of rebuilding the broken characters and then using neural network for recognition. Kannada characters are difficult to recognize due to their complicated shapes and braking of characters makes it much more difficult. The broken characters are therefore rebuilt using end point algorithm to remove brokenness and a single layer neural network is used for classification. A recognition accuracy of 98.9% was achieved for broken characters on synthetically generated data sets.