3D Model Retrieval Method Based on Feature Lines

In order to avoid the limitation of the input sources in 3d model retrieval, this paper proposed a kind of 3D model retrieval method based on feature lines with natural images as the input sources. Firstly, a set of optimal viewpoints was trained and obtained based on the best viewpoint extraction algorithm. Under these optimal viewpoints, mixed outline views of a 3D model were rendered. Then, Gabor edge response characteristics were extracted from these mixed outline views in order to establish a feature library. Finally, the same edge response characteristics of the input natural image were extracted. The similar models were retrieved from the feature library using the bag of visual word method and the retrieved models could be ordered by their similarities. In this method, the edge similarity between natural image and feature line view of the model is used to retrieve 3D model. It can reduce the interference of degenerative view and natural image texture on 3D model retrieval. It corresponds with the mechanism of human visual recognition on 3d objects and this method has a good retrieval effect.