Habitation Extraction And Edge Optimization

Automatic extraction for habitation and optimization for outer profile in panchromatic images can deal with the difficulty in land objects recognition as well as the large intension in manual extraction work; Meanwhile, the processing takes an important role in objects classification, image analysis and objects recognition, and so on. The work mentioned in this paper that focuses on extraction for habitation and optimization for outer profile are showed as follow.1. The paper brings forward a method of statistic standard based on finite threshold segmentation about habitation texture in middle or small scale remote sensing images, this way inducts six statistic standard based on finite threshold segmentation: image mean value, image variance,change frequency in x/y direction, grain number, mean value of grain area. Selecting different threshold in fixed texture area, then aiming at six statistic standard on segmentation the analysis work is implemented.2. By applying SOM method, the result of statistic standard on segmentation on different threshold is used to be input sample for SOM following certain rule, the maps model are trained to extract habitation. Then, clustering processing is implemented, and the habitation blocks are gained.3. Concerning the requirement of mapping synthesis and the algorithmic character of the paper, basing on changing threshold interval of texture feature of images which are extracted and steps of clustering, five rules have been drawed. Deleting holes, fixing area threshold, eliminating one-pixel protrude on four corners, filling one-pixel hollow and one-pixel hollow on the edge, then extending N-pixels protrude or hollow on edge straightly to improve habitation polygon, and the optimized process has been achieved by the means of morphology.By applying the habitation extraction method presented in this paper, it is no use imparting seed point to every habitation. In case feature maps have been trained, automatic processing is used to deal with images with same period and sensor, without processing habitation blocks in every image, which makes the habitation extraction work convenient, speedy and efficiently and reduces the manual work to a great extent. The habitation boundaries extracted by the method are quite regular, and can be easily optimized. Through the edge optimization rules and methods designed in this paper, exterior habitation boundaries can be effectively extracted.