An efficient and effective hybrid pyramid kernel for un-segmented image classification

Automatic object annotation usually requires complicated segmentation to separate foreground objects from the background scene. However, the statistical content of a background scene can in fact provide resourceful valuable information for image retrieval. In this paper, we propose a new hybrid kernel that incorporates local features extracted from both dense regular grids and interest points for image classification, without requiring segmentation. Features extracted from dense regular grids can better capture information about the background scene, while interest points detected at corners and edges can better capture information about the salient objects. In our algorithm, these two local features are combined in both the spatial and the feature-space domains, and are organized into pyramid representations. From the experimental results, we observe that our algorithm achieved a 4.5% increase in performance as compared to other pyramid-representation-based methods. The proposed hybrid kernel has been proven to satisfy Mercer's condition and is particularly efficient in measuring the similarities between image features. For instance, the computational complexity of the proposed hybrid kernel is proportional to the number of features.