采用几何特征的人造物体图像检索研究

Research on Image Retrieval of Artificial Object Based on Geometric Features

修正传统图像信息描述方法,在人造物体图像解释过程中采用几何特征和先验知识相结合的方法,提出人造物体图像结构描述以及相应的匹配算法。通过特征点对目标图像的形状轮廓进行离散化和曲线拟合获取离散曲线段的几何特征向量,并根据特征点类型来描述曲线段之间的关系,最终还原人造物体图像结构描述的参数化模型。实验证明,该方法更能够满足人类的视觉特性,而且能够方便计算机还原出图像中的对象形状。

To qualify the method of traditional description of image, this paper combines geometric features with prior knowledge in artificial image recognition process, and proposes the method of structure description of artificial object image and the corresponding matching algorithm. It makes the graph structure of object discretization by feature points, and obtains its geometric characteristics with curve fitting. Then it describes the relationships of curves by the types of feature points, and develops the parametric model to restore the object in the image. Experimental results show that the description method has better human vision and can help computer easily restore the shape of the object in the image.