Text Regions Extracted from Scene Images by Ultimate Attribute Opening and Decision Tree Classification

In this work we propose a method for localizing text regions within scene images consisting of two major stages. In the first stage, a set of potential text regions is extracted from the input image using residual operators (such as ultimate attribute opening and closing). In the second stage a set of features is obtained from each potential text region and this feature set will be later used as an input to a decision tree classifier in order to label these regions as text or non-text regions. Experiments performed using images from ICDAR public dataset show that this method is a good alternative for problems involving text location in scene images.