Image Representation with Bag of bi-SIFT

Local features are widely adopted to describe visual information in tasks for image registration and nowadays the most used and studied feature is SIFT (Scale Invariant Feature Transform) for the great local description power and the reliability with different acquisition condition. We propose a feature that is based on SIFT features and tends to capture larger image areas that can be used for semantic based task. These features are called bi-SIFT for their resemblance with textual bigrams. We tested the capability of the proposed representation with Corel dataset. In particular we calculated the most representatives features through a clusterization process and used these value according to the "visual terms" paradigm. Experiments on the representation of sets of images with the proposed representation are shown. Although preliminary the results appear to be encouraging.