Remote Sensing Image Segmentation Algorithm Research That Is Based On Histogram Thresholding Method

Along with the rapid growth of remote sensing technology and geographic informationsystem, the high-resolution remote-sensing images were widely applied in agriculture, forestry,national territory, earthquake disaster estimate, military and other domains, and they had broughthuge economic and social returns. The remote-sensing image segmentation refers to be based ona certain division criterion to divide the images into many target sectors or Target objects thatwere interested in. In view of the histogram of image may well present the background regionsand target sectors of the images had big differences in gradation, color and other aspects, usingthe histogram threshold method to divide the images was always a research hot pint ofremote-sensing image segmentation domain. The traditional histogram threshold segmentationmethod was usually used for small-size ordinary image to segment images, the quantity ofnumerical calculation quantity is small and easy to operate.Remote sensing image contains ahuge amount of data and complex structure, using traditional methods are difficult to determinethe threshold, image segmentation effect is not very ideal, limiting the histogram threshold valuemethod is widely used in remote sensing image segmentation. Therefore, how to improve andoptimize the histogram threshold method, and to apply it into remote-sensing imagesegmentation was a very important study content.This article referenced traditional image segmentation method, using the Visual C++andGDAL to construct a remote-sensing image segmentation algorithm that was based on thehistogram threshold method, and applies this algorithm into presenting the SPOT-5remote-sensing image segmentation process in the two-peaks histogram and in three-peakshistogram, divided the water region image information and farmland region image,theexperiment effect is good. The main resourch work are as follows:(1) Constructing a display algorithm for remote-sensing images. The algorithm is based onVisual C++for platform, combined with the GDAL library to read and display the remote-sensing images, and supporting a variety of image formats. The Implemented functionsincluding: identify image waveband range, display under the Gray scale Model, display differentvalues combination of RGB waveband under color mode, and so on.(2) Proposing a new method to constructe remote-sensing image waveband histogram, thismethod using Visual C++and GDALRasterBand correlation function. To construct the remote-sensing image wave band histogram through establishing a quantification coordinatesystem in advance, to draw the waveband histogram curve in the coordinate system.(3) Using the automatic threshold selection method to detect the peak and Valley values ofSPOT-5image histogram, to compare Digital Number and the local area of its neighboring byreferencing, determined the threshold of histogram segmentation.(4) This article using Visual C++with the open source library GDAL to construct a remotesensing image histogram threshold algorithm. The algorithm can classify the image pixels toimplementation image segmentation according to histogram threshold. By taking advantage ofthe algorithm the SPOT-5images which presented as the shape of bimodal histogram is studiedand the water information of the image is extracted. The accuracy is92.31%. Combined with themethod of numeric sieve which can eliminate binary image spots, the extraction accuracyreached94.5%after segmentation processing water extraction diagram. The segmentationaccuracy is increased by about2.2percentage points compare with the method of the histogramthreshold value.(5) In this study, several methods are contrasted to each other such as the binarysegmentation method, fixed threshold segmentation and watershed algorithm, and realized on theplatform of VC++. In addition, making reference tothe Maximum Likelihood Method andMahalanobis Distance Method, accuracy of the water extraction are86.67%,73.33%,85.25%,90.91%, and85.25%. The experimental results show that accuracy of the histogram algorithmbuilt in this study for water body extraction is the best.(6) Using histogram threshold segmentation algorithm, to present three peak shape ofSPOT-5image segmentation, extract the farmland and water thematic information. On visualinterpretation, to extract water body figure with farmland results comparing with the originalimage and superposition, the overall segmentation effect is good. After verification, extractionaccuracy is91.33%.