Study Of Technologies On Texture-based Image Retrieval In Remote Sensing

With the development of space technology, remote sensing technology, network and database, the data of remote sensing image that we can get is blooming in an astonishing speed. How to scan quickly and retrieve target efficiently in the Region of Interest (ROI) from large remote sensing image database becomes the bottleneck of information extraction and share based on remote sensing image. The technique of Content Based Image Retrieval (CBIR) provides a kind of powerful tool which can extract remote sensing ROI information automaticly, and it also becomes the hot spot at home and abroad. As one of the fundamental visualized characters, texture feature is used widely in the Content-based Remote Sensing Images Retrieval.Considereing the current status and problem of the texture-based image retrieval in remote sensing, the paper has researched on the following aspects:1. This paper expounded and sumed up systematically the key techniques of CBIR, including content-based feature description and feature extraction, and computation similarity of feature and the evaluation of CBIR performance etc.2. It presented the retrieval technique of texture spectrum histogram according to the concept of the texture element. Based on this technique, it proposed a kind of improved texture spectrum, and used the Hu invariant moments and histogram invariant moments to extract texture feature from texture spectrum images, which can improve the image invariant of revolution. It was proved by the experiment that the method of improved texture spectrum invariant moments has better precision than ordinary method of texture spectrum.3. It had researched on the characters of using wavelet decomposition to extract and retrieve texture feature according to the property of wavelet decomposition. By analyzing wavelet decomposition and the distribution of the texture feature in wavelet coefficient, it introduced a gradual retrieval scheme of tree wavelet decomposition, which can make up the scarcity of high frequency in texture feature in the proceed of pyramid wavelet decomposition, and realized the retrieval of texture feature from raw to precise.4. It analyzed and researched the block-based data organization strategy in the technique of texture-based image retrieval. According to the fact that the retrieval of remote sensing image is the retrieval of ROI, therefore, the retrieval of region texture feature can be realized by the block-based data organization of large frame image texture. In the consideration of precision and efficiency of texture-based image retrieval, the data organization strategy of quin-tree was recommended. And multi-texture feature retrieval can improve the precision of retrieval.