关键词: 图像配准, 互信息, 部分体积插值, 局部极值, Blackman-Harris窗sinc函数

Abstract: In algorithm of medical image registration based on Mutual Information（MI），when the translation component is integer times of pixel size，conventional PV（Partial Volume） interpolation method will result in the emergency of the local extremes in mutual information registration function，which may hamper the optimization algorithm from getting accurate match parameters.An improved PV interpolation method is proposed by using Blackman-Harris windowed sinc function as kernel function.In addition，the number of concerned neighborhood pixels increases to 16 from 4.Local extremes are eliminated effectively and smooth MI function curve is acquired.The experiments show that the new method is feasible and has higher robustness.

Key words: image registration, mutual information, partial volume interpolation, local extremes, Blackman-Harris windowed sinc function