基于非下采样剪切波变换和特征合成的医学图像融合算法

针对融合后的医学图像时常存在细节纹理不够清晰的问题，本文提出一种新的基于非下采样剪切波变换（Non-Subsampled Shearlet Transform，NSST）的医学图像融合算法，对多模态医学影像进行融合，增强细节结构提取的能力，提高图像融合质量，为医疗诊断提供依据.首先，将已配准的源图像进行NSST分解，得到低频子带和一系列高频子带；其次，对于低频子带系数，提出利用局域平均能量与局域标准差的合成值进行子带之间选择的融合策略，有利于完整保存基础信息，对于高频子带系数，利用改进的拉普拉斯能量和（New Sum of Modified Laplacian，NSML）的方法进行融合；接着，将融合过后的低、高频子带进行NSST的逆过程变换，从而得到融合之后的图像；最后，在灰度和彩色医学多模态图像上进行大量的实验，并选择信息熵（IE），空间频率（SF），标准差（SD）和平均梯度（AG）对融合后的图像进行质量评价.仿真结果表明，本文算法在主观视觉效果以及客观评价指标上均取得较大改善.与其他算法相比，信息熵，标准差，空间频率和平均梯度的平均值分别提高了2.99%，4.06%，1.78%和1.37%，融合后的图像包含更丰富的细节纹理信息，视觉效果更好.

Medical Image Fusion Algorithm Based on Non-Subsampled Shearlet Transform and Feature Synthesis

Aiming at the problem that the detailed texture is not clear enough for the fused medical image, this study proposes a new medical image fusion algorithm on the basis of non-subsampled shearlet transform (NSST) to fuse the multimodal medical image to enhance the detail structure extraction, improve fused image quality and provide a basis for medical diagnosis. First of all, the registered source image is decomposed by NSST to obtain a low-frequency sub-band and a series of high-frequency sub-band. Then, for the low-frequency sub-band coefficients, this study proposes a fusion method using sub-band selection between the regional average energy and regional standard deviation. For high-frequency sub-band coefficients, the fusion method is performed using the new sum of modified Laplacian (NSML). Afterwards, the fused low-frequency, high-frequency sub-band coefficients are inversely transformed by NSST to obtain a fused image. Finally, a large number of experiments were performed on grayscale and color medical multimodal images, and IE, SF, SD, and AG were selected to evaluate the fused images. The simulation results show that the proposed algorithmimprove subjective visual effect and objective evaluation. Compared with other algorithms, the average values of IE, SD, SF, and AG increased by 2.99%, 4.06%, 1.78% and 1.37%, respectively. The fused image contains more detailed texture information and better visual effect.