| With the improvement of global remote sensing satellite technology and drone photography technology,the spatial resolution of acquired images is getting higher and higher.How to accurately segment various types of ground objects from high-resolution images is an important research direction.This paper focuses on the watershed algorithm in image segmentation algorithms.The specific research contents of this article are as follows:(1)Four types of segmentation algorithms are analyzed.It also introduces the basic principles of the watershed algorithm and the basic principles of the algorithms related to this paper.Aiming at the shortcoming of over-segmentation in watershed algorithm,this paper proposes two types of improvement of watershed algorithm.First,modify the gradient map through various operations,that is,modify the target image before running the watershed algorithm;second,perform region merging on the segmentation results of the watershed algorithm according to certain rules.Both directions can reduce the over-segmentation problem of the watershed algorithm.(2)An improved watermarking algorithm based on threshold is improved.First,the threshold is based on a thresholded watershed algorithm.The algorithm first uses median filtering in the RGB color space to suppress noise in the input image.Secondly,the Sobel edge detection operator is used to calculate the gradient image of the input image.Combining mathematical morphology and Otsu to modify the grayscale image,the foreground label image is obtained by the regional maximum value of the grayscale image,and the foreground label is modified according to certain rules.Then use the distance watershed algorithm to obtain the background mark map,and use the mark map to modify the gradient map,and finally use the modified image to perform watershed segmentation.(3)An improved marker watershed algorithm combined with region merging is improved.The algorithm first uses the median filter to process the RGB three color components of the input image,and also uses the Sobel edge detection operator to calculate the gradient image of the input image.Then use the mathematical morphology to process the grayscale image,obtain the foreground mark by obtaining the regional maximum value of the grayscale image,obtain the background mark by the distance watershed algorithm,then modify the gradient map according to the mark,and perform watershed segmentation using the modified image.Finally,the segmentation results of the watershed algorithm are combined according to certain rules.(4)Image segmentation experiments were performed using Google satellite images and drone images.Using two improved algorithms and the classic labeled watershed algorithm to segment Google satellite images,the experimental results show that the two algorithms in this paper do not over segment the water bodies in satellite images 1 and 3 and the roads in satellite image 2.The classic algorithm has obvious over-segmentation.Threshold-based labeled watershed algorithm and classic labeled watershed algorithm were used to perform segmentation experiments on drone images.The experimental results show that algorithm 1in this paper does not exist for segmenting water bodies in UAV image 1 and roads in UAV image 2.The phenomenon of over-segmentation is particularly serious. |
