Remote Sensing Image Segmentation Based On Watershed Algorithm Research Of Problem

With the development of sensor technology and the improvement of theperformance of computer, the resolution of the remote sensing image is gettinghigher and higher, and the data of the remote sensing image is increasing moreand more. The continuous development of remote sensing technology canimprove the rate of information extracted and the accuracy of informationmonitored, which it also brings difficulties and challenges for remote sensingimage analysis. Combining the segmentation algorithms of computer vision andthe characteristics of remote sensing image, the studying of the segmentationalgorithm which is more suitable for remote sensing image is becoming a hotresearch. The requirements of object-oriented segmentation ideological forimage segmentation, which shouldn’t only consider the spectral characteristicsof the pixel, but the spectral characteristics and the spatial structurescharacteristics of pixel should be considered. An effective solution is acombination of edge-based segmentation algorithm and region-basedsegmentation algorithm of computer vision.There is a serious problem for conventional watershed algorithm that isover-segmentation. There are two solutions for this problem:1. reduce theproduction of over-segmentation factors before watershed transform;2. proposerules for region merging after watershed transform. This paper uses the firstsolution given to reduce the production of over-segmentation factors. First filterout the noise algorithm effectively, then reconstruct on the mark image based onmathematical morphology algorithm to reduce the production of over-segmentation factors.Traditional linear filter will blur the details of the characteristics of theimage signal much seriously when filter out some noise, especially for thetexture-rich images. Nonlinear filter will maximize their high-frequency detailsof the images when filter out some noise, which is widely used in filtering. Theproposed algorithm uses morphological filter based on multi-scale contourstructure elements, which can filter out the noise effectively and retain the edge information, then uses multi-scale multi-directional structural elements whichcan enhance the ability of the edge extraction in the calculation of the gradientimage.The improved morphological filter doesn’t filter out of the dark noise andirregular texture of the gradient image. The proposed algorithm uses extendedminimum transformation for mark extraction, then reconstruct on the markimage after filtering out the noise. This paper will give an effective and easyalgorithm to selection the parameter, and the proposed algorithm uses efficientiterative algorithm to select the most suitable parameter h.Experimental results show that the algorithm can effectively solve theover-segmentation problem of the watershed algorithm. The next further studywill focus on how effective each direction edge detected will be fused ratherthan simple weighted. When algorithm uses the mark and reconstruction to solvethe over-segmentation problem, the next step will focus on the introduction ofthe area of the region of local minima and the ratio of area to the depth of theregion of local minima to further improve the performance of the algorithm.