两种基于灰度相似性测度的超声波图像配准方法的比较

图像配准是图像分析和处理的基本问题，其在医学影像分析、遥感遥测、计算机视觉等领域有着广泛的应用。为了能稳定、准确地进行超声波图像配准，基于灰度信息的提取、变换，提出了以下两种基于灰度相似性测度的超声波图像配准方法：第1种方法是利用Harris角点检测方法提取特征点，然后由特征点提供灰度信息，其配准中的相似性测度定义为一个评价函数（cost-function）。误差评价函数；第2种方法引用了相同的评价函数，但使用的有关唯一性控制和区域对应规则是与第1种方法不同的。在给定了相似性测度的情况下，参数化的超声波图像配准可以表述为最小化的问题。第1种方法还利用多项式映射的方法来变换整幅图像，并估计了其平方和误差。实验的结果表明，这两种算法都很稳定，且合乎图像配准的要求，仅仅是第2种方法比第1种方法的性能要好一些。

A Comparison of Two Similarity Measures in Intensity-based Ultrasound Image Registration

Image registration is a basic problem during image analysis and image processing. Image registration has been widely applied in the fields of medical image, remote sensing and computer vision. This paper investigates two registration methods for ultrasound image relying on the intensity-based similarity measure. In tiw first method intensity information is provided by feature points which have been extracted using Harris corner detector. The registration similarity measure is then defined as a cost-function-error cost function. In the latter method the same cost-function is employed, but uniqueness control and region correspondence are different from the first method. Given this similarity measure, parametric ultrasound image registration is stated as a minimization issue. The experiment even exploits polynomial technique to transform the whole image dataset and estimates the sum of square error in the first method. Image quality is one of the most important issues in ultrasound imaging. Ultrasound images tend to be noisy due to the presence of speckle. Speckle originates from the constructive-destructive interference of the ultrasound pulses, and is produced to some extent by all ultrasound machines. This paper shows that it is possible to perform accurate registration on data from in-vivo examinations. The result indicates that the two methods are robust and meet our requirements, but the latter outperforms the first one. It can match B-scans and areas with certain angel, severe deformation and even with plus noise.