Application Of Superpixel Segmentation Method In High Resolution UAV Image Information Extraction

Image segmentation as the foundation point of the computer vision is irreplaceablein the field of image processing status.The result of Image segmentation directly affect the precision of the subsequent series of work.As the basis of some characteristics of image information,making whole image segmentation into small areas according to the differences between the pixel information collection.At last,The purpose is to simplify the image for the following work.High resolution remote sensing image rapidly occupy so many industries since produced.Because of the much more complexity than low resolution images,there is still not a segmentation method can be widely used.This limits its further development in a certain extent.So it is very necessary to find a appropriate method.In this paper,introducing the idea of super pixel division,aimed to make the high resolution remote sensing image segmentation more efficient.The super pixel here is in terms of pixels,actually,super pixels is a series of meaningful collection of pixels.This method will make a picture of a pixel leveldivided into regional level.To some extent,this method of super pixels is to reduce the complexity of the follow-up image processing,and get super pixels keep the edge information of objects in the image better.Based on the data of unmanned aerial vehicle image,integrating the concept of super pixels to study the segmentation method by the following work.(1)Analysis of the segmentation algorithm which hinders the further application of high resolution remote sensing image.According to three categories of segmentation method widely used in several typical algorithms,Implementing the principle and effect by unmanned aerial vehicle(uav)images and then making summarizes that include the advantages and disadvantages of different methods.(2)Changing the pixel into areas by the method of super pixel segmentation.To improve the working efficiency and reduce the complexity of the task,utilization of the characteristics of the image information.To improve the working efficiency and reduce the complexity of the task by keeping the target edge information on the basis of comprehensive utilization of the characteristics of the image information.This paper mainly study watershed algorithm and summarize the principle and the existing problems.In order to improve the existing problems in the algorithm,combining with the method of mathematical morphology.After watershed segmentation,using the method of normalized to cut image spectral and shape information as the main basis to achieve region segmentation.(3)Designing an experiment that combined watershed segmentation?morphological filtering and normalized segmentation three methods together to extract the segmentation cultivated land plots.At last,evaluating the result of the implementation by accuracy rate and recall rate.It shows that super pixel combines all advantages of the three methods,to some extent,it has made the land parcel accurate and efficient extraction and reduce the error due to subjective factor.