Abstract:
Recognizing village names in handwritten Tamil presents a significant challenge for those working in the postal sector, due to variations in handwriting. This research aims to address this issue by developing and evaluating four different approaches. The first approach utilizes Histogram of Oriented Gradients (HOG) feature extraction combined with a Support Vector Machine (SVM) for classification. The second approach implements a Convolutional Neural Network (CNN) for the direct recognition of handwritten village names. The third approach utilizes the CNN for feature extraction and SVM for model training. The fourth approach integrates HOG and CNN for feature extraction and SVM for classification. Each of the 30 village names considered for this study is represented by 250 images, resulting in a total of about 7,500 images. SVMs with HOG features achieve a recognition rate of 92.11%, while CNNs achieve a recognition rate of 91.45%. The CNN - features combined with the SVM model yields a recognition rate of 95.61%, and the hybrid CNN-HOG features with the SVM model achieves a recognition rate of 98.25%. These results indicate that the hybrid CNN-HOG based feature extraction approach with the SVM model outperforms both the SVMs and CNNs.
