High Resolution Remote Sensing Image Segmentation Method

With the continuous improvement of the technological level, the resolution of theremote sensing image is getting higher and higher and it’s used more and more widely.Compared to the low-resolution remote sensing image, high-resolution remote sensingimages can describe surface features of ground targets more clearly. It’s a hot topic inresearch that how to make full use of the rich information of high-resolution remotesensing images to let it more conducive to practical application. For high-resolutionremote sensing images, traditional analysis methods, which are based on the pixels, cannot be effective anymore. Thus, the researchers proposed object-oriented approach.Image segmentation is the key step in object-oriented analysis method, which isalso the pinch point. By image segmentation, the image is divided into a lot of the initialregions, which are the basic units of the object-oriented analysis of remote sensingimage. Thus, image segmentation has a great influence on the subsequent processing.The scene of high-resolution remote sensing images is complex; there are many regionsthat have different characteristics and close spectral information. Then traditionalsegmentation methods are difficult to conduct high-precision image segmentation,which will incorrectly segment a number of regions in the image. Thereby, it willadversely affect the subsequent processing.In high-resolution remote sensing images, ground targets have clear details andrich texture information. The paper focuses on how to describe the texturecharacteristics of high-resolution remote sensing images. Local Binary Pattern caneffectively describe image texture information. To take advantage of that, textureinformation is used in the image segmentation to classify different regions, which canimprove the accuracy of segmentation.It’s not effective that using the traditional operator to classify the different regionsof the high-resolution remote sensing image. This paper defines an improved rotationinvariant uniform mode Local Binary Pattern operator, the ability of which to describetexture features of high-resolution remote sensing image is stronger. Experiments canprove that the ability of classifying the different regions that in the high-resolution remote sensing image is stronger. Applying this operator in the method, we can easilyclassify those regions, which have close spectral information and different textureinformation, prevent the different areas is divided into the same region and improve theaccuracy of segmentation.This paper presents two merging criterions of texture regions, which is based onregional Local Binary Pattern value distribution and is respectively based on theKullback-Leibler distance and Bhattacharyya distance. In this paper, I add these twomerging criterions to improve the Statistical region merging algorithm, take fulladvantage of high-resolution remote sensing image information, greatly improving theaccuracy of segmentation. These two algorithms can flexible control imagesegmentation scale and they have better segmentation results than ENVI5.0andstatistical region merging method.