Research On Remote Sensing Image Retrieval

With the rapid development of contemporary technology, especially for the techniques of remote sensing data acquisition and the great progress of network, it has received vast amounts of image data since decades; However, choosing some images you are interested from a large number of remote sensing images collect or database, has become a very tedious thing. The common image query methods, at present, mainly depend on the text retrieval, which is an extremely mature search approach. Actually, this manner that adds the text labels for images, which is maybe very ambiguous, cannot meet the needs of the people in image retrieval. Indeed, a new search method should be carried out urgently. Content-based Image Retrieval (CBIR) is a quite dynamic way currently, the kinds of retrieval methods has been developed from the initial pixel-pixel type of the raw data retrieval, to the current feature-based image retrieval, which is increasing popular queries research nowadays, and the highest level of CBIR, semantic-based image retrieval, is so exciting and looking forward to. In this paper, it is involved in the method of the feature-based level. This stage of image retrieval related to the essence of the image, which is the reason why an image is different from other images. There are several original properties or fundamental characteristics:such as color histogram, brightness, shape factor, texture feature, spatial frequency spectrum graph and so on. With the help of these feature vectors, image retrieval can be completed effectively on the basis of images visual information processing and analysis from the low level to high. Content-based image retrieval method has good retrieval performance, and CBIR is the future direction of development of image retrieval.As one kind of raster image data, remote sensing image has the same properties as the general images. Certainly, remote sensing image retrieval can learn from content-based image retrieval to increase the efficiency of remote sensing image retrieval; meanwhile, there are many special features in remote sensing images which are not found in general features, such as spatial information and spectral information. Taking these characters as the parameters in the process of feature matching, I believe, can improve the accuracy of image retrieval results greatly. In addition, content-based image retrieval technique combines remote sensing image processing, image database technology, computer vision and other technologies. The study on content-based remote sensing image retrieval will stimulate the advance use of remote sensing image data, so this technical research has important realistic significance.This paper focus on the content of the key technologies of remote sensing image retrieval research, in this paper, a solution on image retrieval, combining the color feature with spectral feature in remote sensing image retrieval, is proposed. The key to this solution is extraction of features and similarity calculation. We have adopted the HSV model to compute color features, and Normalized Difference Vegetation Index, Normalized Difference Barren Index, Normalized Difference Water Index, three spectral features are adopted to participate the image similarity computation in Euclidean distance formula. In this paper, we extracted the color features vector and spectral features vector of images by using Matlab 2008,ENVI 4.7 and ERDAS IMAGINE 9.1. Based on Geodatabase model of ESRI, a database of features vector and relevant remote sensing images was constructed. Finally, under the environment of Visual Studio 2005, using C sharp programming language and ArcGIS Engine 9.2 component technology realizes the function of image retrieval. The experiments show that the method of using color feature and spectral feature in RS image retrieval is effective.