Arabic Handwritten Word Category Classification Using Bag of Features

Human writing is highly variable and inconsistent, and this makes the offline recognition of handwritten words extremely challenging. This paper describes a novel approach that can be employed for the offline recognition of handwritten Arabic words. Through conceptualizing each word as single, inseparable objects, the proposed approach aims to recognize words in accordance with their complete shape. This paper describes the bag-of-visual-words method that has been effectively employed for the purposes of classifying images. The study consisted of four main stages. First, a set of image patches were sampled for the purposes of training, and a speeded up robust features (SURF) descriptor was then used to characterize them. Following that, the bag-of-visual-words model was employed through constructing the K-means clustering algorithm. A histogram of each whole world was developed and this operated as the image feature vector. This was employed to train the support vector machine classifier, which was then able to effectively distinguish between handwritten words. Finally, the effectiveness of the proposed method was tested using a sample of Arabic words extracted from the IFN/ENIT database and the results indicated that the bag-of-visual-words approach represents a promising method of recognizing and classifying handwritten Arabic words. The best and average recognition rates of the proposed method are 85% and 75% respectively.