Content-based 3D model retrieval using relevance feedback

The relevance feedback aims to solve the subjective similarity elimination gap between high-level concepts and low-level features. It first was proposed in the text retrieval, and later investigated and developed in content-based image retrieval. Though relevance feedback has been a lively topic of research in text retrieval and in image retrieval, it has hardly been explored in 3D model retrieval, which focuses on the geometry feature's the selection, the extraction and the matching, scarcely using people's visual subjectivity. This paper presents a novel heuristic relevance feedback algorithm. During the feedback process, users need only to mark relevant or irrelevant model according to their respective information needed, then retrieval system tries to retrieve very similar models to the user's query via studying. The relevance feedback (RF) technique makes effective improvement on content-based 3D model retrieval performance. The method processes 3D model with texture, and extract texture feature.