Feature Extraction Of Buildings In High-resolution Remote Sensing Images Based On Moment Invariant Algorithm

With the rapid development of remote sensing satellite business, especially for thelaunch of a series of satellites with high-resolution sensors such as IKONOS,WorldView and QuickBird, etc., the information acquired by remote sensing imagesincreases largely, which has greatly promoted the advancement and application ofremote sensing data in much area. Constructions are regarded as an important groundobject target and the feature extraction and identification of them is very significant inthe construction of digital city, urban planning and military reconnaissance, etc. Howto accurately extract the feature of constructions has been the research hotpot. In thispaper, we proposed to introduce moment invariant algorithm into the area ofconstruction feature extraction, and then we did many experiments to confirm theeffects of this method. The specific work was as followed:1）This paper summarized the current research on construction extraction forremote sensing data. There is so much research on construction extraction forhigh-resolution remote sensing data and the research on matching by moment invariantalgorithm in the digital image database is also very common. But, the method usinginvariant moment for feature extraction of remote sensing images is still beingexplored. In this paper, we used moment invariant theory to finish feature extractionfor remote sensing images of construction.2) As we know the edge detection and segmentation process should beimplemented before feature extraction, experiments based on had been done by edgedetection.Also, we chose five suitable methods of edge detection by combining theevaluation index of edge detection and visual analysis. Based on the effect of edgedetection, mark watershed algorithm was used to split image, and the threshold ofsegmentation was set in the process of experiment. By comparing the number ofsegment areas and quantificationally analyzing the information entropy, the bestsegmentation threshold was determined to avoid excessive segmentation and to get theperfect the segment result.3) Moment invariant algorithm was used to conduct feature extraction of construction for the high-resolution images of IKONOS and WorldView with theresult of segment conducted above, which was the main work and unique feature. Forone thing, A number of experiment were implemented to get the range of variation ofmoment invariant, by which threshold could be extracted and the construction featureextraction could be finished. For another, sample figures for evaluating the extractioneffects of algorithms were drawn by specialized image processing software. The finalresult of this research was quantitatively evaluated based on three kind of evaluationstandards: accuracy, error rate and leakage rate. By comparing with other extractionalgorithm, the feasibility and efficiency of the algorithm proposed by this paper wasproved.