摘要： 交叉累计剩余熵(CCRE)比传统互信息在配准强噪声图像时更具优势,但采用部分体积(PV)插值的CCRE在网格点容易产生局部极值,不利于变换参数的优化。针对该问题,研究基于3阶B样条函数的PV插值(BPV)、哈宁窗sinc函数的PV插值(HPV)和Blackman-Harris窗sinc函数的PV插值(BHPV)方法在CCRE中的应用,提出一种新的插值方法。该方法采用灵活的邻域中心,将插值点对联合直方图贡献的权重分散到临近的9个点上,并使用高斯函数作为PV插值的核函数,避免权重突变。实验结果表明,与BPV,HPV和BHPV插值方法相比,该方法对噪声图像的配准率较高,配准速度较快,更适合应用于CCRE的计算。
关键词:
图像配准,
交叉累积剩余熵,
部分体积插值,
高斯函数,
3阶B样条函数,
sinc函数
Abstract: The key strength of the Cross Cumulative Residual Entropy(CCRE) over the popular Mutual Information(MI) method is that the former has significantly larger noise immunity.But CCRE using conventional Partial Volume(PV) interpolation will result in the emergency of the local extremes on grid points,which may hamper the optimization algorithm from getting transformation parameters.In order to solve this problem,three improved PV interpolation methods are studied,including 3-order B-spline PV interpolation(BPV),Hanning windowed sinc PV interpolation(HPV) and Blackman-Harris windowed sinc PV interpolation(BHPV).Meanwhile,a new interpolation method is proposed which uses flexible neighborhood center and
makes the interpolation point to distribute the weight of the joint histogram to its adjacent 9 points.Moreover,it uses a Gaussian function as the PV interpolation kernel function to overcome the mutation of weight.Experimental result shows that the registration accuracy and speed of the proposed method is higher and faster,compared with BPV,HPV and BHPV method.So it is more suitable for CCRE computing.
Key words:
image registration,
Cross Cumulative Residual Entropy(CCRE),
Partial Volume(PV) interpolation,
Gaussian function,
3-order B-spline function,
sinc function
中图分类号: