基于互信息量图像配准中目标函数局部极值的成因分析与克服

图像的配准在医学临床科研、诊断等方面具有重要的应用。基于互信息量的全局图像配准方法具有自动化程度高、配准精度高等优点。但是，当平移距离为象素的整数倍时，插值算法会引入伪影，使目标函数产生局部极值，使得最优化搜索有时会终止于局部极值，得到错误的配准参数。本研究分析了这种插值伪影产生的原因，并提出了克服方法，经实验检验，该方法使目标函数进一步光滑，有效克服了插值伪影，保证了最优化搜索的准确性，提高了互信息量方法配准的成功率。

ANALYSIS AND OVERCOME OF LOCAL MAXIMA OF CRITERIA FUNCTION IN MUTUAL INFORMATION-BASED IMAGE REGISTRATION

Image registration has important utilization in medicine. Mutual information-based global image registration provides high level automation and accuracy, but suffers from the artifacts induced by interpolation resulting in the local maximums. Errors may occur while performing the global optimization where the local maximums reside. The reasons why the artifacts occurred were analyzed and an overcoming method proposed. More function smoothness has been acquired and the success probability greatly improved.