| Building area detection and classification has wide application in constructingdigitized city. On the one hand, building area detection is helpful in the automaticclassification of street view, and it could also be used to set up texture database ofbuildings for the further achieving texture mapping automatically in the establishmentof3D city map. On the other hand, it can be used to reduce the search scope for featurepoints detection, matching and identification, which improve the efficiency of buildingtarget identification and3D reconstruction, it plays a very important role in militarytarget recognition and so on.The traditional building area detection methods mainly include the templatematching method and the building consistency analysis method. The first one has highprecision, but requires a priori knowledge, and the process is more complicated; thesecond method, however, process is quite simple but without enough precision. In orderto improve the precision of the building area detection and at the same time insure theprocess remaining relatively simple, this paper proposed a method based on the gestaltprinciple which is the foundation of human visual perception theory. It goes into themethod of building area detection base on the Gestalt principle. The main works andachievements are as follows:1、In the section3we proposed a building extraction method based on linearfeature set. Linear feature is an important feature of man-made objects; buildings areman-made object so without no doubt they contain a lot of this kind of feature. Thischapter focus on the whole process of getting the structure of the straight lines thatbelonging to the building by detecting the linear features and then screening andclassification. Those straight lines demonstrate the extraction of image area of thebuilding. Firstly, get the straight line segments by the use of phase marshalling;secondly, marshal those obtained line segments by reusing Gestalt rules of proximityand consistency and then get a long straight section set; thirdly, classify those longstraight section sets by using k-means clustering method and then those differentstraight line sets help determine the mass-tone attune of the building. From the maincolor information, the straight lines that do not belonging to the building target can beeasily recognized and eliminated and then comes out the target straight line set. Those full steps finally help get the target building area detection. At the end of the chapter, itsvalidity is shown by the experiment. This algorithm overcomes the limitations oftraditional matching algorithms that require the priori knowledge, at the same time itsolve the problem that the traditional analytical method can only handle the singlescene.2、In the section4we proposed another building extraction method which is basedon the Gestalt principle. In this method, human perception, building outline and itsregional characteristics have been combined skillfully in the building extraction whichhas greatly improved the extracting accuracy. Step one, work out the significance of thebinary edges to obtain the edges with relative strong significance and then make use ofthe iterative search method to determine the bottom boundary of the building in thecomplex scenario; step two, on the basis of the integrity of the building, the method ofgraph model is proposed to extract the closed contours from the target image; and thentake the Gestalt rules on mathematical quantitative calculation and in the meantime bydesigning the edge energy function, we can screen out the optimal closed contours;finally, combined with the features of building area like texture, gray and position, it iseasy to merge and delete the wrong building area and get the accurate target area. Theexperiment shows that, compared with the existing methods, the method recommendedin this section is quite more accurate in the target area extraction and with higherintegrity in the extracted outline, in addition, it can extract multiple targets at the sametime. |
PDF Full Text Request