中文摘要：

针对目前图像融合算法基于像素级实现、未充分考虑图像纹理特征、融合效果不理想的现状,本文通过结构张量特征值构造特征模板,在梯度域通过特征加权获得融合梯度场,充分考虑图像特征在融合过程中的决策力,将特征级融合思想嵌入到算法中,在此基础上实现三维数据的直接融合,同等对待3个维度的信息。在头部PET/CT图像融合实验中,本文算法较基于小波变换的三维数据融合方法,清晰度提升64%,交叉熵提升21%,客观评价优势显著。从图像整体亮度、边缘清晰程度等视觉效果方面比较,本文算法亦优于基于小波变换的融合方法。最后,将融合灰度图像和PET源图像通过Alpha半透明图像叠加进行伪彩色显示,进而提升融合结果中有效信息的辨识程度。

英文摘要：

The fusion results of existing image fusion methods are always dissatisfied for these methods don＇t fully considered the texture character of image in the pexil-level fusion algorithm. By constructing feacture template from eigenvalues of structure tensor, performing feacture weighting in the gradient field and comprehensively considering the features of images,the three-dimensional（3D） fusion is completed, which realizes the direct fusion of 3D data, achieves feature-level fusion and treats the information of three dimensions equally. In the head PET/CT fusion experiment, compared with 3D fusion algorithm based on wavelet transformation, the proposed method has approximately 64% and 21% improvement in clarity and cross entropy. Evaluated by subjective assessment, such as image brightness and edge sharpness, the proposed method performs better than the fusion method based on wavelet transformation. The identification of useful information is improved by using Alpha-Blend pseudo-color fusion method for the fusion grey image and PET image.