Research On Remotely Sensed Images Retrieval With Relevance Feedback Based On Arbitrary Shape

As the resolution of remotely sensed images which are available increases and thecommercial object-oriented image analyzing software prevails, the analysis, recognitionand retrieval of remotely sensed images at the level of object have been becoming animportant part of remotely sensed applications.In this paper two feature extraction algorithms of high-resolution remotely sensedimages based on arbitrary shape are discussed at length. The first algorithm is a directfeature extraction method based on a variety of split shapes(such as cellular), which doesnot require changing the shape, such as “split-and-merge”; The second needs changing theshape of image, including “padding” and “projection”. The “padding” method is widelyused in the present study of the standard texture image retrieval, however,“projection”method is not yet widely applied to image retrieval research. This paper proposes cellularshape, uses “projection” for image retrieval innovatively, and try to use “padding” methodin remotely sensed image retrieval based on arbitrary shape. And through experiments ondatabase composed of multiple types of remotely sensed image objects, the performanceof the feature extraction and retrieval are compared, analyzed and verified.On the basis of studying feature extraction and retrieval algorithms of high-resolutionremotely sensed images based on arbitrary shape, the paper combines relevance feedbackand muti-features by good use of users’ relevance feedback information and remotelysensed images information, in order to improve the retrieval performance. Throughexperimental results and quantitative analysis on the retrieval performance, the feasibilityof relevance feedback methods depicted in this article are fully confirmed andconsolidated. Research results also suggest that the use of muti-features in remotelysensed image data retrieval can effectively compensate for the insufficiency caused byusing single feature and accordingly obtains better retrieval performance.A remotely sensed images retrieval system based on arbitrary shape whichimplements the technique of relevance feedback is constructed. This research hastheoretical and application significance on extraction of remotely sensed information,sharing and broadening remotely sensed image data.