Skip to Main Content
This paper presents an image feature extraction algorithm that enhances the object classification accuracy in the spatial pyramid matching (SPM) framework. The proposed method considers the spatial statistics of the feature vectors by calculating the moment vectors. While the original SPM algorithm captures the spatial distribution of the image feature descriptors, the proposed algorithm describes how such spatial distribution is variant. The experiments are conducted using two state-of-the-art SPM-based methods for two commonly used datasets. The results demonstrates the validity of our proposed algorithm. The cases where the proposed algorithm works well are also investigated. In addition, it is demonstrated that the proposed feature and adding more layers improve the classification accuracy in different situations.