基于纹理上下文的遥感图像目标识别

发布时间：2018-06-02 01:04

本文选题：纹理上下文特征 + 统计流形　；参考：《系统工程与电子技术》2017年10期

【摘要】：基于统计流形理论并结合形状上下文思想,提出了能够描述图像纹理目标的纹理上下文特征,进而实现对遥感图像纹理目标的识别。首先将图像的灰度概率密度函数看作统计流形上的点,用所得到的图像统计流形模型来描述图像的纹理上下文特征;然后使用Fisher信息距离来度量流形上点之间的相似度,并利用匈牙利算法来匹配纹理上下文特征;最后通过计算匹配距离来实现不同图像目标的相似性度量。实验表明,与经典的灰度共生矩阵、局部二值模式和统计流形算法相比,对于具有纹理特征的遥感图像,该方法具有更高的识别率且具有普适性和稳健性。
[Abstract]:Based on the statistical manifold theory and the idea of shape context, this paper proposes a texture context feature which can describe the texture object of the image, and then realizes the recognition of the texture object in the remote sensing image. Firstly, the gray level probability density function of the image is regarded as the point on the statistical manifold, and the image statistical manifold model is used to describe the texture context feature of the image, and then the similarity between the points on the manifold is measured by using the Fisher information distance. The Hungarian algorithm is used to match the texture context, and the matching distance is calculated to measure the similarity of different image objects. Experiments show that compared with classical gray level co-occurrence matrix, local binary pattern and statistical manifold algorithm, this method has higher recognition rate, universality and robustness for remote sensing images with texture features.
【作者单位】：西北工业大学理学院;
【基金】：国家自然科学基金(61301196) 国家级大学生创新创业训练计划(201610699329)资助课题
【分类号】：TP751

【相似文献】