基于稀疏表示超像素分类的肿瘤超声图像分割算法

Tumor ultrasound image segmentation algorithm based on sparse representation of superpixel clustering

目的:图像中目标的识别与分割一直是图像处理的研究热点。本文针对超声图像提出了一种新的基于超像素区域特征的肿瘤识别分割算法。方法:首先利用简单线性迭代聚类算法产生超像素,将图像分为多个内部特征相似的图像块,然后提取每个区域的特征组成该区域特征向量,利用稀疏表示分类算法(Sparse Representation Classification,SRC)构造分类器,对超像素进行分类合并,最终识别并分割出完整的感兴趣区域。结果:本文的算法在超声图像中肿瘤的识别与分割中取得较为理想的效果,灵敏度指数平均值达到了83.79%,标准化的Hausdorff距离指数平均值达到了4.80%。结论:本文的分割算法克服了SRC算法不能获得目标区域完整轮廓的缺陷,并取得了较好的实验结果,为超声图像中肿瘤的识别与分割提供了新思路。

Objective Target recognition and segmentation of image has been a hot research topic in image processing. A new tumor segmentation algorithm based on the superpixel area characteristic in ultrasonic image is proposed. Methods Simple linear iterative clustering （SLIC） algorithm was firstly applied to produce superpixel, dividing the .image into multiple image blocks with similar characteristics. And then, the feature of each region was extracted as feature vector. Classifier was constructed by sparse representation classification （SRC） algorithm, classifying and merging the superpixels. Finally, the complete interesting regions were identified and segmented. Results The algorithm for the tumor identification and segmentation in ultrasound image achieved satisfactory effects. The average value of sensitivity index reached 83.79%, and the average value of normal Hausdorff distance reached 4.80%. Conclusion The proposed algorithm overcomes the disadvantage of SRC algorithm which cannot obtain the complete contour of target region, achieving better experimental results and providing a new thought for the tumor recognition and segmentation of ultrasound images.