Classification in image recognition by ambiguous components

Abstract

In the image recognition field, there are many proposed artificial intelligence techniques for finding features that can differentiate data belonging to different classes. Features or components which appear ambiguous for separating data belonging to different classes are usually left out in this field. In this paper, we will demonstrate that by proper design those ambiguous components can still be used for differentiating data. We proposed an association rules based method for designing an image classifier that can distinguish natural images and text images. Our experiments indicate that when existing approaches fail to carry out correct classification, our method can undoubtedly achieve better resu