Statistical and Similarity Features Based Recognition of Offline Characters

This paper uses a blend of similarity and statistical based features for the recognition of offline alphabetic characters in noisy and noiseless environment. The complete representation of the characters is based on the combination of these two different families of features and recognition by different classifiers. The main strategy is to extract complementary similarity measure (CSM) as a feature vector and combined with grey level co-occurrence matrix (GLCM) features. A standard dataset is taken into consideration and recognition is done by artificial neural network (ANN), support vector machine (SVM), Naive Bayes (NB) classifier and random forest (RF) classifier. The highest average recognition accuracy of all characters is recorded as 94.05% using RF in noiseless environment. In noisy environment, the highest accuracy is recorded as 75.8% by neural network. The analysis proves that the combination of feature works on various types of printed characters in noisy and noiseless environment irrespective of the font of characters.