Amharic Handwritten Character Recognition Using Combined Features and Support Vector Machine

Handwritten character recognition is one of the most challenging problem in the area of pattern recognition. In addition to different writing style, in Amharic character there is visual similarity of shape, and Amharic characters are large in number. This paper addressed the challenges and difficulties of Amharic handwritten character recognition by combined various feature extraction techniques, such as HOG, LBP and geometrical features. LDA is used to reduce the dimensional of combined features. And multiclass SVM using ECOC framework is used as a classifier. The algorithm is trained and tested on Amharic handwritten characters data set and Chars74K benchmark numeric data set. The model is validated using 10-fold cross-validation technique. As a result, Multiclass SVM classification algorithm and combined feature extraction technique achieved good result in recognizing Amharic handwritten characters.