Remote Sensing Image Retrieval Based On Radiation And Spatial Information

Developed along with the modern information technology, Remote sensing technology has provided the basic data and become the key part in the construction of global information technology. As the result of aerospace technology, sensor technology, network technology and database technology development, the quantity of Remote Sensing image obtained has been growing exponentially. This, undoubtedly, creates the most favorable condition for geospatial information to develop quickly and effectively. But, at the same time, People's effective management of the megabit of remote sensing image, for example, organizing, browsing, search and retrieving lags far behind the speed of growth of remote sensing image data itself. This becomes the bottleneck of combination of remote sensing image information and its practical application. Therefore, the research for retrieval of image data has become an important and critical issue the field of information processing technology, which not only with high research value, but also with more extensive practical prospects.In this paper, an image retrieval method based on the combination of radiation and spatial information has been put forward considering the characteristics of multi-spectral, multi-sensors, multiple perspectives, multi-spatial resolution and megabit of remote sensing images. With the method proposed, we can effectively improve the retrieval effectiveness. The prime tasks of this article are as follows:(1) Considering the characteristics of multi-spectral remote sensing images and multi-polarization SAR images, an image retrieval method based on K-L transform is proposed. That is, first, do K-L transform on the multi-dimensional data, using the first few principal components of the raw data to reduce the dimensions in image retrieval, and this has improved the retrieval efficiency.(2) Introduce the spatial information in the retrieval using the piecemeal method. Use the rectangular piecemeal and the annular piecemeal method separately to piecemeal the images, then extraction spectrum characteristic of these sub-block images for image retrieval. Experiment explained that the rectangular piecemeal method is sensitive to the picture rotation, while the annular piecemeal method has the invariability to image revolving.(3) Introduce the spatial information in the retrieval using the method of Extracting interest points. Block method can not solve the problem of image rotation, while the retrieval based on interest points is not only invariant to image rotation, but also to the translation. In this paper, the Harris and SIFT operator were used to extract the points of interest, and achieved ideal results in the retrieval.(4) When extract the image feature, we contrast two methods, that is, the histogram and color moment to statistic the spectrum characteristic, and in this paper we select histogram as statistics of the image spectrum characteristic; When measuring the image similarity, we choose the histogram cross method out of several methods through the experiments.