Ensemble Recognition of Persian Typed Sub-word in limited Space Using Smart weighted voting

In this paper, an ensemble method for recognition of Persian typed sub-words is proposed. First, the search space is limited to a very small number of sub-words using a few simple features. Then, by combining the six basic classifications with the weighted voting method, the sub-word is recognized. One of basic classifiers is the same as the search space limiter. Four of the basic classifiers use the nearest neighbor method with each feature of the loci, zoning, the number of the vertical cross between text and background and DCT, respectively. In another classifier, using the product of the normalized image of the input sub-word and the images of the reduced training sub-words, A degree of similarity is obtained for each training sub-words And with its help, the sub-word is recognized. The final sub-word is selected from the options obtained in a weighted voting process whose optimal weights are obtained by an intelligent algorithm. This method has been tested for lotus font and 98.34% recognition rate has been gained for this data.