摘要
皮肤病变的精确分割是实现皮肤病图像自动分析的关键步骤,为后续的特征提取、病变分类等步骤提供了便捷性。然而,现有方法存在分割不足或分割过度的问题,通常会出现皮损的边缘部分丢失和背景错分的情况。针对这些问题,提出了一种新的自动皮肤病变分割方法,该方法包含4个步骤,首先预处理皮肤病图像,去除毛发噪声,增加分割的精确度;随后利用超像素对图像进行初始分割,形成网格状图像;然后采用密度聚类算法对颜色相近的超像素进行合并;最后使用形态学方法处理得到最终的分割结果。在ISIC2018公开的皮肤镜图像数据集上进行实验,结果表明,所提算法与其他分割方法相比:分割结果更精确,更鲁棒,另外从分割指标上也可得出,该分割算法在边缘处理上更加完美。
Accurate segmentation of skin lesions is the key step to realize automatic analysis of skin disease images,which provides convenience for subsequent steps such as feature extraction and lesion classification.However,the existing methods have the problems of insufficient segmentation or excessive segmentation,which usually results in the loss of skin lesions and the wrong separation of background.Aiming at these problems,we propose a new automatic skin lesion segmentation method which consists of four steps.Firstly,the skin image is pretreated to remove hair noise and increase the segmentation accuracy.Then the image is initially segmented by superpixel to form a mesh image.Next the superpixels with similar colors are merged by the density clustering algorithm.Finally,morphological methods are used to obtain the final segmentation results.Experiment on the dermoscope image data set published by ISIC2018 shows that compared with other segmentation methods,the proposed algorithm is more accurate and robust in segmentation results.In addition,it can be concluded from the segmentation index that the proposed segmentation algorithm is more perfect in edge processing.
作者
赵晓梅
刘兆邦
张正平
谢璟
陆千琦
ZHAO Xiao-mei;LIU Zhao-bang;ZHANG Zheng-ping;XIE Jing;LU Qian-qi(School of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China;Suzhou Institute of Biomedical Engineering and Technology,Chinese Academy of Sciences,Suzhou 215163,China;Wenzhou People’s Hospital,Wenzhou 325699,China)
出处
《计算机技术与发展》
2020年第6期167-171,共5页
Computer Technology and Development
基金
浙江省自然科学基金(Q19H110002)
温州市科技计划项目(S20170011)。
关键词
皮肤病变
超像素
密度聚类
合并
形态学方法
skin lesions
superpixel
density clustering
merge
morphological method
