Improving Character Recognition Accuracy with Meta-Heuristic Algorithms

Abstract

Background

Recognition of symbols and words is crucial in today's digital age, with Artificial neural networks algorithms (ANN) playing a significant role in this domain. The primary challenge addressed in this research is the need for a reliable and efficient system capable of achieving high accuracy in character recognition, despite varied font styles and minimal training data.

Materials and Methods:

Our study demonstrates combining Artificial Neural Networks with two Meta-Heuristic include Grasshopper and Propeller algorithms significantly improves accuracy of character recognition system. Many pre-processing techniques applying in order to achieve optimal segmentation of these character. After that, twenty-seven statistical features such as geometric, shape and size are extracted for capital and small alphabet character. Back-Propagation Learning Algorithm (BP) was used for training the ANN, optimising its performance and fine-tuning internal parameters over 1200 iterations.

Results:

This hybrid approach achieves high accuracy, more than 93% in both capital and small alphabet characters. The evaluation algorithms give 0.90 %Sensitivity and 0.93%Specificity.

Conclusions:

According to evaluation algorithms, the combining Artificial Neural Networks with two Meta-Heuristic algorithms achieves high accuracy character recognition.