Remote Sensing Image Thresholding Segmentation Based On The Otsu Algorithm

As a key technology of remote sensing image processing, remote sensing image segmentation is one of the hottest spots and focus of research scholars, having a wide range of applications.Usually, remote sensing images are presented as more gray scales and informatives, fuzzy boundaries, complex structure and so on.When the traditional Otsu algorithms are used for remote sensing image segmentation, the segmentation results are not ideal. As a result, the following of remote sensing image analysis, understanding and recognition are affected. To solve this problem, although there have been many improvements of the Otsu algorithm, but most of which are with one-sided of a design, not a strong versatility. In particular, when these improved Otsu algorithms are used for remote sensing image segmentation, the effect is still poor. So, facing with many different types of remote sensing images, how to improve the Otsu algorithm and its applicability of remote sensing image segmentation is worthy of study.In this paper, the shortcomings and deficiencies of the Otsu algorithm are detailed analysised based on the basic principles of the Otsu algorithm. After which, the constraints of gray-scale Otsu algorithm and the fast recursive Otsu algorithm based on the scatter-difference are proposed. Meanwhile, in view of the noise contained in remote sensing images, the fast recursive Otsu algorithm based on the scatter-difference is extended to the two-dimensional space.The experimental results show that the proposed fast recursive Otsu algorithm based on the scatter-difference has good versatility and adaptability to different types of remote sensing images. Especially when it's extended to the two-dimensional space,the algorithm not only has better segmentation results, but also has a certain noise immunity and high efficiency. Therefore, this paper completes the work of improving the Otsu algorithm and its adaptability in remote sensing image segmentation,the fast recursive Otsu algorithm based on the scatter-difference algorithm proposed in this is an effective method of remote sensing image segmentation, having some practical value in engineering practice. The innovative contributions of this research paper are presented as follows:1. Discover and prove the conclusion that the Otsu segmentation algorithm's threshold is the divided two types of object and background mean average. On this basis, conclude that when the difference between the two types of class variance is large, Otsu segmentation algorithm's threshold will favor a large variance between classes.This conclusion provides a theoretical basis of how to determine the optimal image segmentation threshold.2. In order to slove the problem of Otsu threshold shift, a constraints of gray-scale Otsu algorithm which uses the second search to find the optimal segmentation threshold is proposed.This algorithm greatly reduces the threshold shift and increases the remote sensing image segmentation accuracy.3. A modified Otsu algorithm based on scatter difference and its fast recursive algorithm is proposed in order to solve the problem of computing large, ignore the variance within classes and threshold shift.The experimental results show that this algorithm improve both the accuracy and the speed of remote sensing image segmentation.4. In view of the actual situation that remote sensing images often contain noise, the fast recursive Otsu algorithm based on the scatter-difference is extended to the two-dimensional space.Not only ensuring the accuracy and the speed of remote sensing image segmentation, but also having a certain anti-noise ability.