Research On Multi-source Remote Sensing Image Correlation For Retrieval

With the widely appliance of massive detectors like remote sensors and aerialsurveyors, countries all over the world are attaining data of Multi-source Aero SpatialImage Information that are increasing in a geometrical progression. People have arrivedthe big-data times. So how to organize these data and to support decisions duly andscientifically has become the focus of researches on image mining and retrieval.There are some evident deficiencies in the existing researches especially when itcomes to practice. First, they prefer to the optimization retrieval technology rather thanpretreatment of data for fear of the large quantity, as a result, the research cost is tooexpensive to accept and the outcome is not so satisfying. Second, the correlation rulesare probably concentrating on the link of texts like semantics or contents like colors ortexture, there are also some researches that try to combine the earmarks of texts andsemantics, but the earmarks of the images’ intrinsic coordinate information hardlyattract the researchers’ attention. Third, the designs of retrieval systems are confined in“query by texts” or “query by contents”, amalgamation of images’ multifariousearmarks is seldom considered.To conquer the deficiencies mentioned above, the paper provides a pretreatmenttechnology for retrieval named “Building of correlation Graph Model” that availablyamalgamates the textual information, the content information and the coordinateinformation together.First, with the help of metadata catalogue model, the text information such as thename, the sort, the platform, and the sensor parameter; the content information such asthe color, the venation, and the shape; the space information such as the coordinate andthe range are distilled and organized perfectly.Next, on the base of metadata information, the paper pre-calculates the textualcorrelation between two images by ameliorated classical arithmetic, pre-calculates thecontent-correlation by the technology of mining correlation rules basing on content, andpre-calculates the space-correlation by an arithmetic proved in this paper, which canjudge the spatial relation in topology, in direction and in distance. Then syncretizes thetextual, the content and the spatial earmarks, and a dependent weight is provided basedon the thought of training data to get the whole unitary weighted correlation and in theend the nodes whose correlations are larger than the threshold prearranged are givenback and the correlation database is built.Finally, basing on the graph model, a method is provided to change the classicalexertive arithmetic to the retrieval of the stare node and child-graphs, which reduces thetime spending effectively and improves the abundance of the results. With the help ofthe prototype system built by the paper and the normative data, some experiments are carried out to testify the dependability and practicality of the arithmetic.