Design of High Efficient and High Recognition Rate for Real Time Handwritten Recognition using HMM and ANN Classification
Description
Handwritten character recognition has been an active and challenging research problem. Most of the traditional methods have two challenges, due to the large variations of characters and the dependency relationship between characters. First, in real applications, words may be written cursively, so it is hard to identify the words automatically. Even if the words are neat, different people may write the same words in different styles. Since there are large shape variations in human handwriting, recognition accuracy of handwritten words is very difficult. Finally, dependency relationship between characters has not been well utilized. For these reasons handwritten character recognition has become a highly active and challenging research problem. To recognize the handwritten words, in the proposed work the hybrid mode that combines Artificial Neural Network (ANN) and Hidden Markow Model (HMM). Due to the large variations of words, a feature extractor which could obtain transnationally invariant features is necessary. ANNs are well suited for handwritten character recognition problems. The output of each layer of ANN can be regarded as features directly, but the features maybe non-linear separable.