An Image Segmentation Method Based on Peak-valley Principle and K-means Algorithm
Wenjie Yao, Taihui Liu · 2019
In order to avoid K-means algorithm falling into local optimum solution, this paper proposes an image segmentation method based on peak-valley principle and K-means algorithm and apply it to the Computed Tomography(CT) image. Firstly, based on the observation of a large number of medical image gray histograms, peak-valley principle is summarized. Then, under the guidance of this principle, a K-means algorithm improved by Peak-Valley Principle (PVK-means) is proposed. Assuming that the number of clusters is K, PVK-means algorithm uses neighborhood valley-emphasis Otsu algorithm to select K-1 global thresholds according to the quantitative principle in peak-valley principle. According to the shape invariance principle, the adjacency principle and the maximum principle in the peak-valley principle, the maximum value in the interval is selected as the initial clustering centroid. Finally, K-means algorithm is carried out with the selected initial clustering centroid. The experimental results show that PVK-means algorithm can not only avoid K-means algorithm falling into local optimum solution, but also improve the segmentation efficiency by more than 25%.