Shape Feature Based Multi-class Classification Approach towards Odia Characters employing Extreme Learning Machine

Abstract

Character recognition of Odia alphabets using computer-aided techniques has become a challenging research issue due to its complexity. Odia is recognised as one of the classical languages. Though various image processing methods have been used for classification or Odia character recognition but still there is scope for improvement. The multi-class classification demands the implementation of an elevated constructive learning algorithm. In this paper, we have proposed a conjunctive approach of shape-based feature extraction and Extreme Learning Machine (ELM) to classify the Odia alphabets. The proposed method is implemented over 1500 Odia alphabet images comprising of 52 classes. ELM brings an integrated learning domain with extensive feature transformation which will act as a catalyst for effective fulfillment of classification purposes in the multi class domain. ELM based technique is tested for different activation functions and the output result shows the effectiveness of ELM classifier over traditional Naive Bayes and support vector machine (SVM) classifier. The ELM based technique gives more promising results in comparison with the above two classifiers for the multi class handwritten Odia alphabet classification.