Application Research Of Multi-scale Image Segmentation Based On Improved Otsu Algorithm

Image segmentation is one of the most important technology in digital image processing.So it is always the focus of the image research. Scale is the most basic feature of remote sensing image, and more and more researching work has been done on it.The contradiction of Image Segmentation always exist in the dilemmas between the segmentation accuracy and the segmentation easiness in single scale.Because of the images possess features at multiple scales,so we can combined the multiscale analysis method and the image segmentation method to unify the segmentation accuracy in finer scale and the segmentation easiness in coarser scale.In this paper,we use the IKONOS panchromatic image with a resolution of lm and the Geoeye-1panchromatic image with a resolution of0.5m as data source.Then we combined the multiscale analysis method and the image segmentation method to research the segmentation effect of the new method.The results obtained by this paper are as follows:(1)We introduced a multi-resolution model.In this model,we use local average method to realize the scale expansion from high resolution to low resolution,and we build up a parent-child link by computing parent-child distance to realize the sale contraction from low resolution to high resolution,so we can get the segmentation result at the original resolution.(2)We introduced several common used methods of best scale selection:local variance method and variogram method,then in this paper, we proposed a new method:best scale selection of remote sensing image based on information entropy.In this study,we use the local variance method to calculate the best scale of experiment l,and the result is4.0m. We use the new method to calculate the best scale of experiment2,and the result is1.0m.(3)We combined the multi-scale analysis method and the image segmentation method, and proposed a new method:multi-scale image segmentation based on improved Otsu algorithm.The traditional Otsu algorithm has a disadvantage:threshold shift.So we considered both of the mean and variance functions to improve the traditional Otsu algorithm.At the same time,we introduced a recursive Otsu segmentation algorithm in this paper.We used two types of image segmentation quality evaluation criteria:regional consistency, regional contrast.Through two experiments, we proved that the new method is better than the original method. Finally,we used the multi-resolution model to realize the segmentation results that corresponds to the original resolution.