Partial Relevance Feedback for 3D Model Retrieval

Relevance feedback (RF) proved an effect way to improve the precision and recall of 3D model retrieval. Unfortunately, through existing methods of RF, it is straightforward to find out whether a model is similar or not, but it is impossible to find out which local part is similar or not. The new partial method of RF proposed in this paper provides a good solution, in which not only the similar models are marked out but also the local parts which are similar or not are pointed out and taken advantage at the same time. This additional information contributes a lot to the improvement of 3D retrieval. Experiments show superiority in effectivity of the new method.