基于方差和显著性特征的超声图像分割方法研究

Ultrasonic Image Segmentation Based on Variance and Saliency Features

针对超声图像对比度小导致的影像上相邻灰度差别很小,人眼有时难于区分的问题,提出了一种基于方差和显著性特征的超声图像分割方法。首先提取图像中已知样本像素点的方差和显著性特征;然后利用支持向量机根据提取的样本像素点方差和显著性特征进行样本训练,得到分类模型;最后根据训练模型对整幅图像上的像素点进行分类,实现图像的有效分割。实验结果表明该方法针对超声图像的分割是有效的。

According to the low contrast of ultrasound images can result the difference of adjacent grayscales is very small, and the human eyes are difficult to distinguish, an ultrasound image segmentation method based on variance and saliency features is proposed. Firstly, the variance and saliency features of the known sample pixels in the image are extracted, then the sample training is performed based on the variance and saliency features of the extracted sample pixel points using a support vector machine to obtain the classification model, finally, the training model is applied to the entire image to achieve effective segmentation of the image. Experimental results show that the proposed method is feasible and effective for the segmentation of ultrasound images.