Files

Abstract

In this paper we propose a method for classifying regions of images and videos frames into text and non-text regions using support vector machine (SVM). Different features are proposed to characterise the texture formed by text characters and background. SVM has an advantage that it is insensitive to the relative numbers of training examples in positive and negative classes. This advantage is are illustrated by comparing results with those obtained using a multiple layer perceptrons (MLP).

Details

Actions

Preview