基于平移不变剪切波变换的医学图像融合

Medical Image Fusion Based on Shift-Invariant Shearlet Transformation

传统的基于剪切波变换的融合方法在融合图像的奇异处易产生伪吉布斯现象.为此,文中提出了一种基于平移不变剪切波变换的医学图像融合新方法.该方法利用平移不变剪切波变换将待融合的医学图像分解成低频子带和高频子带,并使用基于区域系数绝对值和权重的规则融合低频系数;对于高频子带,提出一种基于支持向量值激励的自生成神经网络的高频子带融合策略;最后利用逆平移不变剪切波变换重建融合后的图像.视觉比较和融合结果量化分析表明,文中提出的方法不仅能够有效克服伪吉布斯现象,而且以熵、互信息、平均梯度和QAB/F作为标准时其融合结果比小波、轮廓波和非下采样轮廓波变换等传统方法更好.

In the conventional shearlet transformation-based image fusion methods,there commonly exists a pseudo-Gibbs phenomenon at the singularities of the fused image.In order to solve this problem,a new fusion method of medical images is proposed based on the shift-invariant shearlet transformation.In this method,source images are decomposed into lowpass and highpass sub-bands via the shift-invariant shearlet transformation.Then,the lowpass coefficients are combined by employing the scheme based on the region coefficients' absolute values and weights,and the highpass sub-bands are merged by adopting a fusion scheme based on the support vector value-motivated self-generating neural network(SGNN).Finally,the fused image is obtained via the inverse shift-invariant shearlet transformation.Both the visual comparison and the quantitative analysis show that the proposed method effectively avoids the pseudo-Gibbs phenomenon and outperforms the conventional wavelet-based,contourlet-based and nonsubsampled contourlet-based methods in terms of entropy,mutual information,average gradient and QAB/F.