基于综合特征分析的多源空间数据相似检索

在遥感不同波段、不同分辨率、不同时相、不同尺度和不同平台等情况下,观测获得的多源空间数据集成的数据库中,提取感兴趣的局域目标图像是海量数据库管理的关键问题之一。本文提出一种统计法的纹理描述与频谱法的纹理描述相结合,纹理特征和空间关系相结合的方法进行多源空间数据相似检索。统计法纹理描述采用共生矩阵的描述符,而频谱法纹理描述采用树状小波分解(小波包)的纹理描述符。通过对多源遥感数据库的检索比较,表明此方法比单一纹理特征方法具有更高的应用价值。

Similarity Retrieval of Multi-source Spatial Data Based on Combination Features

In this paper, a new approach based on combining two radically different texture features is proposed for similarity retrieval of multi-source remote sensing database. One takes a statistical approach in the form of gray level co-occurrence matrix (GLCM), the other takes a signal processing approach with tree-structured wavelet transform or wavelet packets. Particularly, we also combine texture feature with spatial relation in the statistical method. The similarity retrieval contains six consecutive stages: preprocessing the images in the multisource remote sensing database, feature extraction, feature sequence normalization, weights for feature elements, similarity measure, and experimental evaluation. Through comparison with other methods: such as the gray level co-occurrence matrix, the tree-structured wavelet transform or wavelet packets, in combination GLCM with special relation, the proposed method shows good tradeoff between retrieval effectiveness and efficiency. The experimental results indicated that the combined texture retrieval way has powerful practical merits.