Comparative study of feature extraction and classification techniques for printed bilingual Gujarati-English text

Accuracy of any Optical Character Recognition system (OCR) depends on extraction of optimal features from the text object. To extract optimal features, an appropriate feature selection method needs to be selected from available methods. The character images can be represented perfectly by employing the appropriate feature selection method. This article presents the comparison of two feature extraction techniques along with widely used classifiers. Here, the performance is evaluated based on feature extraction techniques regarding character image classifications. The purpose of the current research work is to identify and show the most optimal feature extraction technique for printed bilingual documents. Here, two feature extraction namely Discrete Cosine Transform (DCT) feature and zone based pixel density feature are used. The classification accuracy is compared and evaluated with each feature extraction method for different classifiers. From the experiments and analysis, it is observed that energy based Discrete Cosine Transform (DCT) feature is outperformed as compared to zone based pixel density feature.