Combining Low and High Level Features for Object Recognition

Object recognition algorithms usually identify: 1) pointbased features and 2) global structure/geometric shape of an object to solve the recognition problem. In this paper, we propose a framework that integrates these two approaches and brings forth the advantages of both to achieve higher recognition rates. In our framework, the object features are represented using biologically motivated C2 descriptors [1], whereas the global structure information is represented with the skeleton structure of the object silhouette. A novel mechanism that integrates the information derived from different object silhouettes to produce a generic skeleton is implemented. We show that the integrated recognition approach using low level (C2 features) and high level object descriptors (skeletons) is robust under occlusions, for variability within specificity and is invariant to geometric changes. The concept of looking at these two disjoint approaches in a common framework is the first such attempt.