基于图像几何形状特征的CBIR算法

CONTENT-BASED IMAGE RETRIEVAL BASED ON IMAGE GEOMETRIC SHAPE FEATURES

张量尺度是一种基于图像几何形状的特征描述子,由于其特征提取过程计算复杂度较高,不适合于快速的基于内容的图像检索。提出一种基于图像森林变换的张量尺度特征提取快速算法,并采用归一化的张量尺度方向直方图作为图像几何形状的特征描述子,与相似性度量标准结合,实现了一种具有图像平移、旋转、尺度变换不变特性的基于内容的图像检索算法。与现有的张量尺度计算方法相比,该算法具有较低的计算复杂度,仿真实验结果证明算法的有效性。

Tensor scale is a feature descriptor based on image geometrical shape,it has a high computational cost in its feature extraction process and is not suitable for content-based fast image retrieval.In this paper,a new fast extraction algorithm of tensor scale descriptor based on image foresting transform is proposed,in it the normalised histogram of tensor scale direction is taken as the feature descriptor of image geometric shape,and by combining the similarity metrics standard,a content-based image retrieval method with invariant properties of image translation,rotation and scaling transforms is implemented.Compared with current methods of tensor scale calculation,the proposed approach has lower computational cost.Simulation experimental results demonstrate the validity of the proposed algorithm.