Digit Recognition Using Hybrid Classifier

This paper proposes a new hybrid classification technique for recognizing printed digits. The feature extraction was performed using object region boundary analysis, Fourier Descriptors (FD) and Chain code based algorithm. A new curve tracing Chain code based algorithm (CTCC) was proposed to extract the curve features from the digit images. The recognition was performed using dynamic programming and multi-layer perceptron using back propagation algorithm (MLP-BP). Higher accuracy of 99% was obtained. The proposed methods were simple with higher recognition accuracy and lesser time complexity.