基于KFCM与多特征融合的皮肤镜病变区域提取  

Extraction of Dermoscopic Lesions Based on KFCM and Multi-Feature Fusion

作　　者：栗碧悦 侯俊[1] 王子硕 薛渊 摆惟圆 

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海

出　　处：《建模与仿真》2023年第5期4595-4604,共10页Modeling and Simulation

摘　　要：病变轮廓的检测是病灶提取、定性研究和探明病灶与其周围组织间关系的基础。针对皮肤镜图像病变区域灰度强度不均匀导致病变区域难以提取的问题,本文提出一种基于模糊核聚类算法(KFCM)与多特征融合的病变区域提取算法。首先,将原始图像进行模糊核聚类,把得到的聚类结果作为多特征融合轮廓提取模型的初始轮廓,其次利用轮廓曲线演化对图像全局信息与局部信息采用自适应加权,构建符号压力函数(Signed Pressure Force, SPF),最后利用轮廓曲线演化来分割图像,提取出病变区域。通过真实皮肤镜图像验证了提出的模型在SPF函数和自适应函数引入两方面均能提升提取性能,且在定性和定量对比评价上均优于其他方法,同时对噪声具有鲁棒性。The detection of lesion contour is the basis of lesion extraction, qualitative study and the relation-ship between lesion and surrounding tissue. Aiming at the problem that the lesion area is difficult to extract due to the uneven gray intensity of the lesion area in dermoscopic images, this paper proposes a lesion area extraction algorithm based on fuzzy kernel clustering algorithm (KFCM) and multi-feature fusion. Firstly, the original image is fuzzy kernel clustering, and the clustering result is used as the initial contour of the multi-feature fusion contour extraction model. Secondly, the global and local information of the image is weighted by adaptive contour curve evolution, and a Signed Pressure Force (SPF) function is constructed. Finally, the contour curve evolution was used to segment the image and extract the diseased area. Through real dermoscopic images, it is verified that the proposed model can improve the extraction performance in both SPF function and adaptive function, and is superior to other methods in qualitative and quantitative comparison evaluation, and is robust to noise.

关 键 词：病变提取 KFCM 活动轮廓模型 符号压力函数 全局和局部信息 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]