FAQ

Q1: Are you talking about multiple feature CBIR systems? It seems that the multiple viewpoint framework is similar to use multiple features to retrieve.

A1: Y/N. A CBIR system can employ multiple features and combine each feature's similarity to get a synthetic retrieval result. This can be regarded as a special case of application of multiple viewpoint technique. Multi-viewpoint retrieval is a high-level concept which more focused on external implementation of a system, while multi-features retrieval is a low-level concept which more focused on internal implementation.

---

Q2: Ok. I agree for a human committee, it is necessary to combine different experts’ opinion, since none of them are perfect. But for computing system, there is no restriction of knowledge or technique on individual system. A non-perfect system just means the technqiues used is not perfect. So why not build a perfect viewpoint instead of combining multiple ones?

A2: It is generally hard to build a complex and powerful system directly. It is relatively easy to build simple systems and weak viewpoints. Incrementally integrating weak viewpoints to build a powerful one is a cost/effective approach.

---