皮肤黑素细胞肿瘤图像综合分割方法研究  

Research on the segmentation algorithm for dermoscopy image

作　　者：田庆飞[1] 孟如松[2] 姜志国[1] 谢凤英[1] 赵丹培[1] 

机构地区：[1]北京航空航天大学图象处理中心,北京100083 [2]全军皮肤病研究所空军总医院,北京100036

出　　处：《中国体视学与图像分析》2007年第3期207-211,共5页Chinese Journal of Stereology and Image Analysis

基　　金：国家自然科学基金项目(No.60672152)

摘　　要：目的本文针对黑素细胞肿瘤(Melanocytic TumorMT)图像情况复杂,较难分割的问题,提出了一种综合数字图像分割算法,探讨MT的早期诊断。方法首先应用统计区域融合方法(SRM)实现图像分割成多块纹理一致的区域。然后对图像以HSV彩色空间的H和S分量为特征,使用K均值聚类算法将图像聚为9类。最后,将聚类结果在HSV彩色空间的H和S分量值分别映射到[0,1]区间,再分别对H分量和S分量取阈值,得到最终的边界分割结果。结果对MT图像能够按照其纹理差异将其有效划分为多个区域,较为准确标识出皮损区域。结论综合对多种方法结果的对比,本方法优于传统的大津阈值法、K均值法和活动轮廓法。同时对过去基于SRM的MT图像分割方法进行了改进,在处理复杂MT图像时效果明显好于传统方法。Object This paper proposes a synthetic algorithm for the segmentation of Melanocytic Tumor （MT）, and discusses about early diagnosis of MT. Method Firstly, the algorithm of this paper uses statistical region merging （SRM） to divide MT image into several areas by texture. Then, based on the features of H and S sub-channels of HSV color space, this paper uses K-means algorithm to classify those areas into nine parts. After that, it normalizes the sub-channels and splits the image into two parts separately by the given threshold value 0.6 in H and S sub-channels. In the end, the algorithm combines the two sub-channels into one and gets the final result. Result The algorithm this paper proposes could divide different areas effectively and assist the doctors to diagnose the disease. Conclusion Compared with many algorithms, the performance of this algorithm is much better than OTSU, K-means and snake algorithms, especially for complicated MT images segmentation.

关 键 词：统计区域融合 K均值 HSV彩色空间 黑素细胞肿瘤 图像分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]