Classification of Graphical Images in Textbooks

In order to automate the process of transforming figures in textbooks into tactile format, the types of the figures play an important role in deciding what process to go through. In this paper, we explore the problem of how to train an accurate Support Vector Machine classifier that can determine whether an image belongs to a certain figure type, using various feature extraction methods. We consider three different types of images: bar charts, diagrams and line graphs. We compare accuracies obtained using three different image feature extraction methods: Principal Component Analysis, Generic Fourier Descriptor and oriented filter based method. We also compare these with text features present in images and the combination of image and text features. In our experiments, we obtained 93% to 98% classification accuracy by combining the features extracted using oriented filters and text features.