Study On High-resolution Remote Sensing Image Segmentation And Object Contour Extraction Algorithm

With the rapid development of space techniques and information techniques,the high-resolution remote sensing satellite can supply human with the high accuracy and plentiful digital images.High-resolution remote sensing images make us observe earth changes in a smaller space scales.The importance of high-resolution remote sensing information extraction is in increasing.Now,for study of high-resolution remote sensing image segmentation and object contour extraction has become the key research subject of high-resolution remote sensing application,on this subject,this paper focus on following aspects:(1) This paper analyses the characteristics of high-resolution remote sensing images and the key of high-resolution remote sensing image segmentation and information extraction.And proof the limitations of pixel-based spectral segmentation with experiments.(2) This paper study on the object-oriented multi-scale segmentation algorithm, use eCognition,experimental conducting multi-scale image segmentation on buildings and roads,and get the best-scale parameters about the image.(3) Study on the algorithm of Meanshift segmentation based on confidence,ACM algorithm,the two algorithms used in high-resolution remote sensing image analysis application.Combinatorial experiments were made with different sorts object for extracting object contour,analyze the characteristics of algorithms.(4) With a number of segmentation methods,many experiments,compared the result of multi-scale segmentation,based on the Meanshift segmentation based on confidence and binary segmentation,after statistical analyze extraction of buildings and roads contour information;carry out the evaluation of the statistical results.(5) After comparison this paper show:Object Oriented Multi-scale segmentation could get better results on relatively large buildings contour extraction;Meanshift segmentation based on confidence could do better on relatively small buildings contour extraction;with regard to the roads contour information,binary segmentation is quite suitable.