基于特征线条的三维模型检索方法

3D Model Retrieval Method Based on Feature Lines

为了避免在三维模型检索中对输入源的限制,提出一种以自然图像为输入源、基于特征线条的三维模型检索方法.首先基于最优视点提取算法训练并获取三维模型较优视点集;然后在较优视点集下渲染三维模型混合轮廓线视图,并为各视点混合轮廓线视图提取Gabor边缘响应特征,建立特征库;最后对输入的自然图像提取相同的边缘响应特征,采用视觉词袋方法从特征库中检索相似模型,并根据相似度排序.实验结果表明,该方法利用自然图像与模型特征线视图的边缘相似性实现三维模型检索,降低了退化视图与自然图像纹理对三维模型检索的干扰,符合人类视觉辨识三维物体的机理,具有良好的检索效果.

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.