Semantic Image Classification using Structure Skeleton with Hierarchical Graph Concept

Abstract

Semantic classification is challenging task in the field of image processing. Many researchers have attempted to improve semantic models such as developing more sophisticated models, or generating intermediate representations by making statistic on low level description. However, the methods are rather rudimentary and it does not specific enough for representing the actual meaning. In this paper, we present a technique of the semantic image classification by using the human perception. The structure skeleton is used to combine the object components and image meaning. The feature selection methods are introduced to select the essential features from existing features. We combine a novel concept called the hierarchical representation graph for producing more semantic classification. This concept is formulates on a graph which is captured the relationships among objects in the images. The experimental results indicate that our proposed approach offers significant performance improvements in the interpretation of semantic images, compared, with the maximum of 80.28% accuracy.