Handwritten character recognition and prediction of age, gender and handedness using machine learning

Abstract

Handwritten character recognition and prediction of age, gender & handedness from handwritten documents offers an interesting research problem for researchers as few research carried out on this field. The aim of this research is to investigate machine learning classification algorithm that is used to recognize different writer’s attributes and their handwritten characters. Predicting writer’s identity and recognizing handwritten characters based on mainly three steps: segmentation, feature extraction and classification. In the segmentation step we used edge detection technique for segmenting dataset images using fuzzy logic. Feature extraction methods are described to take decision category of our writers and their handwritings. For feature extraction we used mRMR for feature selection, tortuosity, direction, curvatures and chain code for feature extraction and PCA for dimension reduction. In the final step, we used KNN, SVM and RFC for classification of writer attributes and recognizing handwritten characters. Classification accuracy on QUWI dataset were 89.41% for recognizing handwritten character, 88.28% for age range prediction, 75.90% for gender prediction and 75.11% for handedness prediction for each writer. We have used these classification algorithms to bring out the maximum accuracy rate for predicting age, gender & handedness.